Engineers have created a drone that camouflages itself as a faint, hovering haze when it takes flight.

The "Phantom Twist" prototype drone is a carefully engineered robot that utilizes motion blur to hide in plain sight — opening up a new frontier in low-visibility flight. Scientists say it could reshape how drones monitor wildlife, survey ecosystems and inspect infrastructure.

Most stealth drones try to disappear by changing their appearance, using camouflage paint, transparent plastics or complex optical tricks to bend or scatter light. The Phantom Twist takes a fundamentally different approach by exploiting how humans and animals perceive motion.

The scientists who created the drone outlined their findings in a study presented July 16 at the conference Robotics: Science and Systems in Sydney, Australia.

"Most efforts to hide drones focus on making them look like their surroundings," said first author of the study Michael Rubenstein, associate professor of computing and mechanical engineering at Northwestern University, in a statement. "Instead, we asked whether we could design the drone itself around the way humans perceive motion.”

The result is a machine that behaves more like a spinning fan than a conventional quadcopter. Typical drones have four rotors spinning around a largely stationary body, so your eye can lock onto the frame even as the propellers blur.

Phantom Twist strips this idea down to a single motor and single propeller. The propeller spins one way while the entire body counter‑rotates in the opposite direction with no stationary parts to visually latch on to — spinning 25 times per second, which is too fast for the human eye to resolve clearly.

It's about 10 times less visually perceptible than a conventional quadcopter, the scientists said. In practice, that means observers don’t see a crisp flying machine but a translucent smudge, with its few opaque components blending into a haze.

To get there, the team assigned algorithms to explore thousands of designs. First, they used a computational model to automatically generate about 20,000 drone configurations that could fly stably, at least in theory. Each configuration shuffled a motor, propeller, batteries, circuit board and counterweight into different positions.

Then artificial intelligence (AI) and optimization tools took over, iteratively rearranging those pieces to minimize how visible each design would be from almost any viewing angle — all while respecting the constraints of aerodynamic stability.

The researchers simulated each candidate spinning in midair and composited the images over 100 different real-world backgrounds, from skies to trees to buildings.

To determine the stealthiest designs, they applied a perception model that approximated human vision, with drones that better blended into their surroundings earning a lower visibility score. From there, the team selected the 500 lowest-scoring designs and ran their optimization loop again, squeezing out further decreases in visibility. When the team was confident in the design of a particular drone, they built it.

The final Phantom Twist isn’t a compact block of hardware; it’s deliberately spread out in three dimensions. Components sit at different heights and angles, with empty space between them, so that when the drone spins, the parts don’t visually overlap into a solid silhouette.

That geometry matters because of the way our eyes process images.

"The human eye takes time to accumulate signals, roughly analogous to the exposure time of a camera," said computer vision expert and co-author of the study Emma Alexander. When something spins fast enough, distinct edges smear together, and we perceive a blur instead of a shape.

The researchers' initial uses for the stealthy drones include monitoring nesting birds without startling them, surveying wetlands without scattering flocks of waterbirds, and inspecting aging infrastructure without altering human behavior, they said.

For now, Phantom Twist isn’t a perfect "ghost." The spinning propeller still generates a significant amount of noise and the spider‑web of thin wires and support rods remains partially visible. Rubenstein and his colleagues said in the study they’re already thinking about next-generation iterations that swap in more transparent materials and quieter propulsion systems.

A four-legged robot has learned to change the way it runs while navigating forests, staircases and obstacle courses. — seamlessly switching between a steady trot and a faster bounding gait without instructions from a human operator.

The 100-pound (45 kilograms) robot, called KAIST HOUND, uses cameras and lidar to scan the ground ahead, then selects an appropriate gait and adjusts its movements in real time. In outdoor tests, it crossed a 0.7-mile (1.1- kilometers) university campus route and a 0.2-mile (0.3 km) forest trail strewn with roots, logs and slippery leaves.

The researchers described the robotic framework on July 15 in the journal Science Robotics.

Changing gait

Animals naturally change their gait depending on their speed and surroundings. A dog might trot carefully across uneven ground, for example, before bounding over a fallen branch. Reproducing this adaptability in robots is tricky because different movements are often controlled by separate, highly specialized coding systems, and transitions between them can cause a lag that drives the robot to stumble.

To overcome this issue, researchers developed a special training framework called action pretrained transformer–based reinforcement learning (APT-RL). This is an artificial intelligence (AI) training system that first studies many examples of actions, uses a transformer to understand patterns across those actions, and then improves through rewards and penalties.

The training began with a simple, two-dimensional computer model of the robot. Using trajectory optimization — a technique that calculates physically workable movements for the robot — the team generated 180,000 short trotting and bounding sequences, including the joint forces the robot's legs need to perform. The dataset represented about 15.5 hours of movement but took only around eight minutes to produce.

During reinforcement learning — a machine learning technique where AI learns to make the best decisions by engaging with a particular environment through trial and error — an AI system then learned how to select and modify those skills while negotiating simulated stairs, stepping stones, hurdles, gaps and rough ground.

In digital simulations, the robot dog was not limited to copying its prerecorded movements. It could also make corrections for three-dimensional terrain and unexpected situations, such as jumping over a log — a behavior that wasn't included in the original, flat-ground training data.

Finally, the researchers configured the system to include the robot's depth camera and lidar scanner in the simulation.

In one indoor test, HOUND bounded across an obstacle 2 feet (60 centimeters) high while briefly achieving 9.5 mph (15 km/h). It also jumped down a three-step staircase. The robot generally chose trotting at lower speeds on irregular ground, while bounding became more common at higher speeds or when it encountered larger steps, hurdles or gaps. The AI system that could select either gait performed more consistently across the different simulated environments than the version restricted to trotting or bounding alone.

The researchers suggest the technology could eventually help robots navigate disaster zones or other places inaccessible for wheeled machines. However, the current framework only allows two gait choices and mainly handles forward movement. Rapid turning, sideways motion and other behaviors like crawling remain future goals for the research team.

Engineers have designed a waterproof "diving suit" for cyborg cockroaches that enables the hybrid insects to survive and roam underwater for up to three hours. This function expands the capabilities of cyborg insects and could one day be deployed in disaster zones, according to the team.

A built-in oxygen generator and silicone tubes deliver the gas directly to a cockroach's breathing holes, known as spiracles. The design is adapted for use in low-oxygen conditions as well as submerged environments, the researchers said in a new study published June 29 in the journal Nature Communications.

"Our approach combines a soft waterproof shell with a simple yet reliable chemical oxygen generator," study co-author Shinjiro Umezu, a professor in the School of Creative Science and Engineering at Waseda University in Japan, said in a statement. "This allows the insect to retain its natural mobility while being protected from an environment that it cannot normally survive in."

Cyborg insects are living insects that have been fitted with electronic controllers that guide their movements. Researchers have previously used them in search-and-rescue operations to access and investigate hard-to-reach areas; for example, they were used in rescue efforts after the devastating magnitude 7.7 earthquake in Myanmar in March 2025 that killed at least 3,700 people and injured 4,800 more. The advantage of cyborg insects over tiny robots is that the former employ insects' muscles to move, whereas the latter rely on high-power batteries that consume energy and can run out of steam.

The cyborg insects deployed in Myanmar were developed in the laboratory of Hirotaka Sato, senior author of the new study and a professor in the School of Mechanical and Aerospace Engineering at Singapore's Nanyang Technological University.

Sato has spent more than a decade pioneering cyborg insect technology. He and colleagues hope the new diving suit will extend cyborg insects' operational range to include flooded and partially submerged areas in disaster zones.

The suit consists of a flexible shell, four silicone tubes that attach to the spiracles and a transparent, 3D-printed oxygen tank. To make the tank produce oxygen, the researchers sprinkled manganese dioxide onto a highly absorbent sponge inside the tank. They then injected a small amount of diluted hydrogen peroxide, which breaks down slowly in the presence of manganese dioxide to produce oxygen. Finally, the team sealed the tank with ultraviolet adhesive to prevent leaks.

"The key engineering challenge was to build a system that was small, light and flexible enough for the insect to wear, while still producing enough oxygen for long-duration underwater movement," Umezu said.

The silicone tubes send oxygen straight into the thoracic spiracles, while the abdominal spiracles, which are lower down the insects' bodies, take in the oxygen contained in the suit.

"Our new insect diving suit works like the oxygen tank used by human divers," Sato said in the statement. The silicone tubes can be attached and removed without pain or harm to the insect, the researchers added.

The researchers tested the suit on a cyborg Madagascar hissing cockroach (Gromphadorhina portentosa), which they placed in a water tank and later sent into a plastic tube that simulated submerged and low-oxygen environments.

The suit enabled the cockroaches to roam underwater for up to three hours, raising the prospect that cyborg insects, including locusts and beetles, could one day be used to inspect flooded pipes, drains, tunnels and other hard-to-access places.

Next steps include improving the diving suit to potentially include sensors and a navigation system; and testing the design in simulated disaster environments, according to the statement.

A weird 20-legged machine could change how scientists think about the ideal robot form.

For decades, roboticists have been inspired by the natural world, building machines that resemble humans, dogs, insects and even horses. But new research suggests that the most useful robot body may look less like a human and more like a sea urchin.

The robot has no front or back. Its 20 telescoping legs, each costing $300, radiate from a central body, with a depth camera at each leg tip, leading the researchers to name it Argus, after the all-seeing monster of Greek mythology. This design results in a machine that can move in any direction, stabilize itself after being pushed, cross rough terrain, carry a 10-pound (4.5 kilograms) payload and even climb up walls.

The Duke University scientists who created the robot published their findings May 27 in the journal Science Robotics.

"Watching Argus move is unlike watching any other robot we've worked with," Jiaxun Liu, a doctoral student in Duke's General Robotics Lab and co-author of the study, said in a statement. "The first time we saw it navigate among trees and rough terrain, even under heavy collisions [when someone pushed it], we knew this was something different."

Simulating symmetry

The team arrived at Argus' design after running more than 1,500 simulations of different robot shapes. Instead of asking what animal the robot should resemble, the researchers focused on how symmetrical a machine could be in all directions ‪—‬ a mathematical concept called dynamic isotropy.

The dynamic isotropy score ranges from 0 to 1 and measures how evenly a robot can accelerate its body, or center of mass, in every direction. A score of 1 means a robot can react or move nearly identically in all directions.

"When a robot can accelerate equally well in every direction, it stops needing to face the world in any particular way," Boyuan Chen, director of Duke's General Robotics Lab and co-author of the study, said in the statement. "Forward and backward become the same. Left and right become the same. The whole problem of robot control changes character."

According to the researchers, most robots today — including advanced four-legged robots, humanoids and conventional drones — score below 0.6, meaning they're better at moving or reacting in some directions than others. With its 20 legs, Argus scored a 0.91, close to the theoretical maximum.

To achieve this high score, the team arranged Argus' body around a shape called a regular dodecahedron, a three-dimensional form with 12 pentagonal faces. The arrangement gives the robot a nearly uniform field of view and allows it to move without needing to orient itself the way a conventional robot would.

Chen said that based on these findings, robots don't need to imitate humans or dogs to boost their agility, and instead are designed from deeper mathematical principles.

Releasing the robot

To test whether Argus' design was truly optimal, the team took the robot out on the Duke campus, where it rolled across concrete, grass, dense foliage, soft sand, wet surfaces and bark. It handled obstacles up to 5 inches (12.7 centimeters) tall, kept moving even after three of its legs were broken, and pushed a 3-foot (1 meter) cube while rolling.

Argus is a proof of concept and not the final answer on the optimal robot design, the researchers wrote in the study. Its broader importance may be in how it was designed rather than where or how it can be used in real-world scenarios ‪—‬ it could be a mathematical way to compare different robot bodies and design new form factors from scratch.

"It shows that designing for dynamic symmetry isn't just a theoretical curiosity," Boxi Xia, a postdoctoral researcher at Duke's General Robotics Lab and co-author of the study, said in the statement. "It produces a robot you can deploy in the wild, on uneven ground and in clutter, even in low-gravity settings. It changes what's possible."

Chinese engineers have built a mecha-style robot that can quickly transition from two legs to four while carrying people, resembling the power-loader exoskeletons from Aliens or the utility-style mobile suits from Japanese anime series Gundam SEED.

The robot's developer, Unitree, says the large, humanoid robot is intended for civilian transport. In a promotional video, the robot ‪—‬ called GD01 ‪—‬ walks upright, smashes down a high wall of cinder blocks, and reconfigures itself to stand on four limbs to traverse more difficult terrain.

Unitree representatives say the machine weighs around 1,100 pounds (500 kilograms) with an operator on board and stands nearly 10 feet (3 meters) tall. People can even buy the robot, with prices starting at 3.9 million yuan ($572,000).

The core structure of the machine is a skeleton of titanium alloy and aerospace-grade aluminum surrounded by a carbon-fiber shell.

Unitree calls the GD01 the world's first mass-produced "transformable mecha," and has urged consumers to "be sure to use the robot in a friendly and safe manner," according to the written description for the promotional video.

Mounting the GD01 in its current incarnation isn't the most user-friendly process. In the video, an operator has to awkwardly scale up the leg of the machine to access the cockpit. Interestingly, although the GD01 is being marketed as a manned machine, the initial shots show it being controlled remotely, with no operator in the cockpit.

Unitree is a robotics startup headquartered in Hangzhou, China, and is best known for much more modestly sized humanoid and quadruped robots.

The company manufactures and sells several models ranging in price from $4,290 for the torso-only R1-D to $90,000 for the H1 — a general-purpose robot built with lidar and depth cameras and driven by Unitree's M107 joint motor, a high torque, high-endurance motor with a focus on agility, speed and load capacity.

At a spring gala hosted by Unitree in February, the company's humanoid robots were filmed performing impressive feats of acrobatics, synchronized and break dancing, and complex martial arts routines.

Unitree has not released a full technical paper on the GD01, but it's clear that the mecha builds on the company's experience building quadrupedal robots capable of traversing difficult terrain.

Four-legged models like the B2 are capable of climbing stairs, remaining upright when suffering heavy impacts, and even leaping across gaps. They can also be modified from straight-legged configurations to wheeled models.

Unitree's quadrupeds use multiple sets of fish-eye binocular depth-sensing cameras, which allow the robots to simultaneously view their surroundings from the front, bottom and sides.

According to Unitree's website, the company focuses on "self-researching key core robot components such as motors, reducers, controllers, Lidar and high-performance perception and motion control algorithms."

An autonomous humanoid robot built by the smartphone maker Honor has shattered the human half-marathon record.

Called the Honor Robotics D1, the mech-inspired, vibrant-red bot completed the 13.1-mile (21.1 kilometers) race in 50 minutes, 26 seconds at the 2026 Beijing E-Town Humanoid Robot Half-Marathon, eclipsing the previous human record of 57 minutes, 20 seconds, which was set by Ugandan runner Jacob Kiplimo in March.

Another separate robot that Honor representatives revealed at Spain's MWC Barcelona (formerly known as Mobile World Congress) in March completed the race even faster — in just 48 minutes, 19 seconds. But this humanoid machine wasn't awarded the win because it was remotely controlled, rather than being fully autonomous, and was thus given a 20% time increase penalty.

The result is especially notable when you compare it with last year's race, when the winning model achieved a finishing time of 2 hours, 40 minutes. Both the second and third runners-up in this year's content were also Honor robots, crossing the line at around 51 minutes and 53 minutes, respectively.

Honor's first robot, that it showcased at MWC Barcelona, is 5 feet, 3 inches (1.6 meters) tall, weighs 176 pounds (80 kilograms) and demonstrates 20 degrees of freedom, with human-like, five-fingered hands. This is unlike D1, which was optimized for aerodynamics with simple, blade-like appendages.

Whereas the MWC bot topped out at 2.5 mph (4 km/h), D1 reached a remarkable average speed of 15.6 mph (25.1 km/h) in Beijing.

A critical part of D1's ability to sustain that speed over long distances is a powerful liquid-cooling system developed primarily in-house. It's a self-developed closed-loop liquid-cooling system adapted from Honor's smartphone technology. A pipeline covers core heat-generating components, especially joint motors, with two high-speed micro pumps with flow rates up to 6 liters per minute of water circulation.

Honor has a long history of incorporating compact cameras, sensors and artificial intelligence (AI) chips into electronics, although these developments have been focused on the company's smartphones. The bridge to humanoid robots is most apparent in another technology Honor debuted at MWC: the Robot Phone.

The Robot Phone combines embodied AI interaction with robotic motion and cinematic imaging capabilities. It has a pop-up camera that serves as the device's "head," and it can bop along to music, track motion, and identify sounds.

The D1 that competed in Beijing is equipped with autonomous perception and navigation, as well as an in-house high-dynamic motion system. Company representatives said in a statement that it's capable of providing real-time interactive feedback during motion. This ability is enhanced by light strips and signature gestures, and Honor intends to deploy it for events such as urban night runs and street sports.

By contrast, the robot that Honor debuted at MWC is positioned more as an assistant. Honor representatives said that the robot can serve multiple everyday support roles, like helping with shopping or workplace inspections, or in a more general role as a "supportive companion."

The world of table tennis may be in for a shake-up after Sony's AI division unveiled Ace — an autonomous robot that can compete with expert table tennis players.

Using a combination of high-speed cameras and proprietary state-of-the-art hardware, Ace scored 16 unchallenged points, or "aces," after serving against multiple elite players.

This is the first time a robot has achieved "expert-level play in a commonly played competitive sport in the physical world," Sony AI representatives said in a statement.

"This breakthrough is much bigger than table tennis," Peter Stone, chief scientist at Sony AI, said in the statement. "It represents a landmark moment in AI research, showing, for the first time, that an AI system can perceive, reason, and act effectively in complex, rapidly changing real-world environments that demand precision and speed."

A study detailing how the robot works was published April 22 in the journal Nature.

Where hardware meets software

AI systems have already shown prowess in strategy games, such as Go, chess and role-playing games.

However, moving AI into a robotic body, where quick reflexes are paired with physical movements, can be more challenging. Here, the software and hardware components have to work seamlessly — and in table tennis, where speed and hand-eye coordination are vital, this pairing has to work to win.

"Table tennis is a game of enormous complexity that requires split-second decisions as well as speed and power," Peter Dürr, director of Sony AI in Zurich and project lead for Ace, said in the statement. "This research breakthrough highlights the potential of physical AI agents to perform real-time interactive tasks, and represents a significant step toward creating robots with broader applications in fast, precise, and real-time human interactions."

Ace's strategy builds on Sony AI's previous research on its AI agent Gran Turismo Sophy, with Ace using advanced sensors and high-speed software to perceive its environment. These sensors include nine active pixel sensor cameras that help Ace identify the ball's exact position in 3D space, along with three gaze systems that use mirrors and event-based vision cameras to measure the ball's spin and angular velocity as it moves through the air.

Running these cameras is Sony AI's proprietary AI control system, which is based on model-free reinforcement learning, where an AI agent learns directly from interactions in its environment without making a predictive model first. This technology allows Ace to adapt and make decisions faster, without relying on a preprogrammed model.

Lastly, Ace's robotic body, which includes a swiveling arm with a paddle-like appendage at its end, was created with the company's robotic hardware.

Beating the pros

In April 2025, scientists had Ace play against five elite players (each with over 10 years of experience and around 20 hours of weekly training) and two professional table tennis players (Minami Ando and Kakeru Sone, both in the Japanese professional league). While players in both tiers are skilled at table tennis, professional athletes make their living by playing table tennis, whereas elite players may not have the same caliber to make the sport their livelihood.

Ace won three out of five matches with the elite players and boasted a 75% serve return rate. Its autonomous system also allowed the robot to return unusual shots, such as balls bouncing off the net. It however lost both matches against the pros.

Then, in December 2025, Sony AI had Ace play a series of separate matches in which it competed against two professional and two elite players. This time, Ace beat both elite players and one of the professionals. Company representatives said the robot moved closer to the table edge, had higher shot speeds and launched faster-paced volleys against its opponents.

Given that less than two years ago, Google DeepMind's robotic table tennis robot was defeated by elite players, Ace's victories show how quickly this field of robotics is advancing in a short time.

"Once AI can operate at an expert human level under these conditions, it opens the door to an entirely new class of real-world applications that were previously out of reach," Stone said in the statement.

Could a humanoid robot perform household chores for you? Soon, that might be the case. Chinese robotics company UniX AI has announced the release of Panther, a humanoid service robot designed to help with common household tasks.

The robot ‪—‬ which weighs 176 pounds (80 kilograms) and stands 5 feet, 3 inches (1.6 meters) tall ‪—‬ can operate for up to 12 hours on a charge, its engineers announced in a statement. In addition to releasing a promotional video showing Panther performing chores like making the bed, cleaning countertops and toilets, and cooking breakfast, UniX AI has started commercial deliveries of the model — suggesting it's ready to be deployed in a real-world setting.

"UniX AI is not only building advanced technology and products, we are bringing robots into homes, hotels, public spaces, and security operations, making them reliable productivity tools for society," Fred Yang, founder and CEO of UniX AI, said in the statement.

A stable service robot

To accomplish household chores, Panther uses two of the world's first mass-produced 8-DoF bionic arms, which give the robot joints with 34 degrees of freedom, UniX AI representatives said.

At the end of its arms are intelligent "grippers" that adapt to different shapes and sizes of items that the robot can grab. In the promotional video, Panther is seen holding a plastic spatula when frying an egg and a toilet brush when cleaning the toilet. The robot even washes its hands in the sink at the end of the video, drying off its grippers on a bath towel.

Panther's body can extend vertically by 31 inches (80 centimeters) to access hard-to-reach areas from tall shelves to ground level. It uses binocular vision and a webcam to see its surroundings.

Whereas other humanoid robots rely on legged locomotion and thus have a higher tendency of falling, Panther uses an omnidirectional four-wheel steering and four-wheel drive (4WS+4WD) chassis, allowing it to smoothly maneuver varying indoor environments.

It uses both lidar and ultrasonic sensors to determine its spatial position and moves around 6 feet (1.89 m) per second — roughly the speed of a brisk walk. UniX AI representatives noted in the statement that engineers focused less on building the "most human-like robot" and more on making sure the model could accomplish a variety of tasks with stability and flexibility.

Taking on multiple tasks

In early tests and demonstrations, Panther handled multiple prompts, including making tea, cleaning the kitchen, and operating a washing machine. To do this, Panther runs on UniX AI's integrated technology stack, which includes imitation learning and long-term task planning so that it can perform tasks seamlessly one step at a time.

Engineers designed Panther for a wide range of applications in addition to housework, including assistance in retail, reception, guided tours, elderly care, companionship, research and education.

The price of this model could not be found on UniX AI's website.

Researchers have created artificial-intelligence-powered robots that can navigate all kinds of terrain autonomously and continue moving even when they're severely damaged.

Dubbed "legged metamachines," these awkward-looking bots may reveal insight into human and animal evolution and provide a path for future robots to overcome mobility limitations, the machines' creators say.

The robots were developed as part of a Northwestern University project that aimed to create robots with "athletic intelligence," meaning the ability to adapt to changing terrain on the fly. The robots are a modular combination of building blocks, described as "Lego-like" in a statement. Each piece is a pair of meter-long (3 feet) legs joined by a central, jointed sphere. The scientists described the robots March 6 in the journal PNAS.

"Inside the sphere, the robot has everything it needs to survive: a 'nervous system,' a 'metabolism' and 'muscle,'" lead study author Sam Kriegman, an assistant professor of computer science and chemical, mechanical and biological engineering at Northwestern, said in the statement. "By that, I mean a circuit board, a battery, and a motor. The modules are mechanically simple. They can only rotate around a single axis, but they are surprisingly athletic and smart."

Robotic building blocks

The bots' modular nature allows multiple pieces or "limbs" to be attached to an individual robot, thus altering its shape and movement without inhibiting its ability to march forward through unstructured terrain.

Unlike most other mobile robots, which have rigidly defined structures and tend to adhere to familiar two- and four-legged designs, these metamachines allow for a much broader number of configurations.

This approach may enable researchers to create and study different mobile forms and examine our notions about the evolution of locomotion, according to the study. In experiments, the metamachines already demonstrated modes of locomotion similar to bounding kangaroos or undulating seals.

Although the combination of limbs may look awkward, the robots exhibit an impressive ability to right themselves when they encounter difficulty, even if they're completely flipped over. They can leap over obstacles and even perform acrobatics in midair.

Simulating evolution

These robots' impressive capabilities are possible due to the team's powerful AI, which simulates an evolutionary algorithm that drives natural selection.

At first, the simulation was purely software-based and confined to operations within a computer program. The AI was tasked with creating novel body configurations from the robot's modular pieces, with the goal of creating the most effective combinations for traversing different terrain.

After the AI had tested designs in a virtual environment, discarding those it found unsuitable, the team assembled the best three-, four- and five-legged designs the model had evolved. The machines could cross terrain broken by gravel, grass, tree roots, leaves, sand, mud and uneven bricks, without interruption or human intervention.

The most impressive piece of the metamachine puzzle was the robots' ability to adapt when damaged. The team ran simulations where various configurations suffered breaks or lost whole limbs, but the modules adapted to keep moving.

By compressing billions of years of evolution into a few seconds, Kriegman said, robotic design can quickly progress beyond traditional configurations.

"Evolution can reveal new designs that are different from or even beyond what humans were previously capable of imagining," Kriegman said. "So, we really wanted to study how and why it works. The best way — or at least the most fun way — is to evolve structures in realistic conditions."

Humanoid robots have taken part in the world's first fully autonomous martial arts performance to mark the Lunar New Year. In new footage from the 2026 Spring Festival Gala in Beijing, Unitree Robotics’ G1 and H2 machines can be seen performing complex acrobatics and wielding weapons.

The machines performed several world firsts, Unitree representatives said in a statement, including the first continuous freestyle table-vaulting parkour, the first launched aerial flip with a maximum height approaching approximately 10 feet (3 meters), and a two-step wall-assisted backflip.

The two robot models were also equipped with new dextrous hands to support the rapid switching between different body positions and transitions between tricks, as well as to grapple surfaces and handle martial arts props.

The robots’ performance is a marked improvement on demonstrations at the same festival in 2025, where the machines looked stiff and clumsy, with clunky transitions between different routines.

Company representatives said the improvements result from artificial intelligence (AI) improvements alongside new lidar (light detection and ranging) processing, alongside mechanical upgrades to the robots themselves. They represent an advancement in the field of embodied AI, in which scientists attempt to give robots tools to better understand and navigate the physical world.

The year of the fire horse

"The performance draws its inspiration from a tribute to traditional Chinese martial arts," Unitree representatives said in a technical blog post. "The concept centers on fusing the martial arts spirit with modern technology, using robotic performances to present the power and beauty of Wushu [the Chinese term for martial arts] while conveying the message of cultural inheritance and innovation. It aims to showcase Chinese Kung Fu culture alongside technological progress to the world."

Company engineers had been preparing for the performance since November 2025. Scientists pretrained a stunt-motion model using extensive training data from an array of stunts, which subsequently improved the standard of fine-tuning to get the robots up to speed.

They also upgraded the cluster control platform — a system for coordinating dozens of robots at once that involves network communication, different operating systems, embedded devices and software engineering elements. This allowed for end-to-end automation, from AI-planned choreography planning to the real-time multirobot coordination, with millisecond-level synchronization.

In terms of robotic motion, the engineers ensured that the robots could navigate their environment, which involved a lidar-based localization algorithm that could improve how accurately they positioned themselves. Another pretrained high-speed-movement model then ensured they could track their target positions stably and quickly and complete the desired movement with human-like gaits.

The machines' design also played a role in the improvements. The engineers upgraded the motors to boost maximum performance and then configured them to adapt to the new movements. Examples of specific improvements include higher power density for core joint motors, optimized limb robustness, and more dextrous hands.

Although the martial arts display was designed primarily for entertainment, company representatives said the movements and actions demonstrate real-world utility. For example, the cluster control platform could be a key step in enabling several robots to coordinate within a variety of environments without human intervention.

In a striking display of agility and grace, a humanoid robot named "Adam-U Ultra" has danced its way through a complex, highly energetic routine without faltering, slipping or missing a beat.

In a new video released by the Chinese robotics firm PNDbotics, the bot — which builds upon the company's original Adam robot, unveiled late last year — is seen performing a sequence of complicated steps. The dance is remarkably nuanced and human-like, and it includes complicated arm, waist and feet movements — all perfectly coordinated with accompanying music.

According to PNDbotics representatives, such complicated motion is possible because of the model's 41 independently controllable joints or actuators. These provide Adam with a huge range of flexibility and motion, enabling it to mimic human motion — even in something as challenging as freestyle dancing. Although the video refers to the dance sequence as the Charleston, it's really more akin to a hip-hop dance routine.

All of those actuators are coordinated by PNDbotic's advanced control system and communication architecture. These include an artificial intelligence (AI) computing platform built around Nvidia's Jetson Orin module — a system-on-module (SoM) that combines a CPU, GPU and other components to create a "brain" for AI platforms.

The robot is constantly iterating and improving on things like stability and balance by combining whole-body control and model predictive control systems built on large-scale simulated environments and neural network training.

Adam-U Ultra also includes a sophisticated vision-language-action (VLA) model, an embodied AI system that sees the world, understands instructions and directly controls actions, enabling robots to perform real-world tasks through natural language guidance. Essentially, you can speak to it to tell it what to do and it will respond accordingly. It's backed up by 10,000 samples of real-world behavior designed to help the platform learn and adapt its movement to changing conditions on the fly.

The vision element of Adam's VLA is driven by an Intel RealSense D455 depth vision sensor, which enables precise 3D environment modeling and real-time spatial awareness, as well as a number of lidar and standard cameras.

The fully humanoid robot shown dancing is being developed in parallel with PNDbotics' Adam-U robot, a stationary model designed primarily as a data collection platform. According to the company's website, four fully mobile humanoid robots are also in development, with varying levels of freedom of movement, sensory capabilities and computing power.

One of the keys to Adam's fluid movement is how relatively lightweight the platform is. Even the heaviest model, the Adam Pro, only clocks in at 139 pounds (63 kilograms), while the lightest is a scant 132 pounds (60 kg). All the models stand 5.2 feet (1.6 meters) tall.

PNDbotics claims that Adam can fulfil a multitude of potential roles, ranging from research and laboratory work to medical assistance and rehabilitation scenarios. It says its robots could help train medical personnel in rehabilitation exercises, monitor patient recovery or even collaborate in surgeries.

The company's website also suggests that Adam could be deployed in traditional industrial roles on manufacturing lines, or in service industries as a concierge or receptionist.

A new technology shoots laser beams at drones to charge them in midair, thus unlocking the possibility of "infinite flight."

Representatives from PowerLight Technologies say the system works by pairing a ground-based transmitter with a receiver on the drone, also known as an unmanned aerial system (UAS). The transmitter combines advanced software designed to provide precise control of the laser beam with hardware that can sustain power transmission on a scale of kilowatts — or thousands of watts — according to the company. For comparison, typical lab lasers operate on a scale of milliwatts, or thousandths of watts.

The software allows the transmitter to lock onto "cooperative" targets and actively track them, thus paving the way for the safe transfer of power to the vehicle, even while it's in flight. PowerLight representatives say its transmitter can operate at heights of up to 5,000 feet (1,500 meters).

For context, the operational altitude of the Raven RQ-11B — a hand-launched intelligence, surveillance and reconnaissance drone developed for the U.S. military — is between 150 and 1,000 feet (46 and 305 meters) above ground level, and 10,500 feet (3,200 meters) above sea level. Larger drones can reach altitudes as high as 60,000 feet (18,288 m).

Integrated control software can also link up with software on the UAS to provide telemetry monitoring and control. This interoperability means that data and commands can be exchanged between PowerLight’s tracking and beam control system, and the drone’s onboard avionics and telemetry system. This could be used to send battery charge information to operators on the ground, who could dynamically alter power delivery, and ensure that it's being delivered efficiently while the laser is on always on target.

"We are building an intelligent mesh energy network capability," said Tom Nugent, chief technology officer and co-founder of PowerLight, in a statement. "Our transmitter communicates with the UAS, tracks its velocity and vector, and delivers energy exactly where it's needed. We have now successfully tested the power transmission and tracking algorithms."

Endless flight

A receiver is built into the drone itself. This receiver, which weighs only 6 pounds (2.7 kilograms), captures the laser energy from the ground-based transmitter and uses a laser power converter to turn it into battery power.

Photovoltaic laser power converters operate similarly to solar cells, capturing light and transforming it into an electrical current. However, they are optimized for monochromatic, high-intensity laser light rather than broad-spectrum sunlight.

The receiver also includes an embedded control module. This module continuously gathers real-time telemetry about the drone's position and operation and relays that information to operators on the ground. It's also capable of establishing a two-way light-based communication link that allows it to both send and receive information using optical signals.

The system was developed in conjunction with the U.S. Department of Defense (DoD). It will be integrated into the K1000ULE, a long-range, high-endurance drone developed by Kraus Hamdani Aerospace. The K1000ULE is fully electric and engineered for persistent, long‑range missions in support of the U.S. Navy and Army.

Previous testing has shown the potential for wireless power transfer through lasers. For example, the Defense Advanced Research Projects Agency demonstrated a power transfer of more than 800 watts delivered from a laser 5.3 miles (8.6 kilometers) away).

However, the PowerLight system is a step toward real-world deployment in mobile platforms. Recent testing has proved the viability of the transmitter and receiver, but PowerLight has yet to conduct flight tests, which are scheduled for early this year.

Can you be certain the person talking to you is 100% absolutely not a robot? Soon, you might not be so sure.

For the first time, scientists have built a robot that can move its mouth exactly like a human. This means it avoids the so-called "uncanny valley" effect, where a bot’s actions appear unsettling because they are uncomfortably close to natural — but don't quite meet that threshold.

The Columbia University researchers achieved the feat by allowing their robot, EMO, to study itself in a mirror. It learned how its flexible face and silicone lips would move in response to the precise actions of its 26 facial motors, each capable of moving in up to 10 degrees of freedom.

They outlined their methods in a study published Jan. 14 in the journal Science Robotics.

How EMO learned to move its face like a human

EMO uses an artificial intelligence (AI) system called a "vision-to-action" language model (VLA), meaning it can learn how to translate what it sees into coordinated physical movements without pre-defined rules. During training, the humanoid robot made thousands of seemingly random expressions and lip movements while it stared at its own reflection in the mirror.

Next, the scientists sat EMO in front of hours of YouTube videos showing humans talking in different languages and singing. This allowed it to connect its knowledge of how its motors produced facial movements to the corresponding sounds, all without any understanding of what was being said. Eventually, EMO was able to take spoken audio in 10 different languages and synchronize its lips near-perfectly.

"We had particular difficulties with hard sounds like ‘B’ and with sounds involving lip puckering, such as ‘W’," Hod Lipson, an engineering professor and the director of Columbia’s Creative Machines Lab, said in a statement. "But these abilities will likely improve with time and practice."

Many a roboticist has tried and failed to create a convincing humanoid, so before unveiling EMO to the world, it needed to be put to the test in front of real people. The scientists then showed videos of the robot speaking using the VLA model, and two other approaches for controlling its mouth, to 1,300 volunteers, — alongside a reference video demonstrating ideal lip motion.

The two other approaches were an amplitude baseline, in which EMO moved its lips based on the loudness of the audio, and a nearest-neighbor landmarks baseline, in which it mimicked facial movements it had seen others make that produced similar sounds. The volunteers were instructed to choose the clip that best matched the ideal lip motion, and they chose VLA for 62.46% of cases — compared to 23.15% and 14.38% for the amplitude and nearest-neighbor landmarks baselines, respectively.

Robot carers will require friendly faces

While there are differences across genders and cultures in how people distribute their gaze, humans in general rely heavily on facial cues when interacting with each other. A 2021 eye-tracking study found that we look at the face of our conversation partners 87% of the time, with roughly 10 to 15% of that time focused specifically on the mouth. Other research has shown that mouth movements are so important that they even affect what we hear.

The researchers believe that overlooking the face’s significance is part of the reason other projects have failed to create convincing robots.

"Much of humanoid robotics today is focused on leg and hand motion, for activities like walking and grasping," Lipson said. "But facial affection is equally important for any robotic application involving human interaction."

As AI technology continues to advance at a breakneck pace, robots are expected to take on an increasing number of roles that require direct interaction with humans, including in education, medicine and elderly care. This means their efficacy will correlate to how well they can match human facial expressions.

"Robots with this ability will clearly have a much better ability to connect with humans because such a significant portion of our communication involves facial body language, and that entire channel is still untapped," said lead author of the study, Yuhang Hu, in the press release.

But his team is not the only one working on making humanoid robots more lifelike. In October 2025, a Chinese company released a video of an eerily realistic robot head, created as part of their effort to make interactions between people and robots feel more natural. The year before that, a Japanese team unveiled an artificial self-healing skin that could make robot faces look human.

When Elon Musk talks about robotics, he rarely hides the ambition behind the dream.

Tesla's Optimus is pitched as an all-purpose humanoid robot that can do the heavy lifting on factory floors and free us from drudgery at home. Tesla is targeting a million of these robots in the next decade.

But is Musk likely to succeed? A few years ago, the thought of a friendly, capable household robot belonged in science fiction. We could imagine machines that danced, shifted boxes or played chess, but not ones that understood us well enough to be genuinely helpful. Then came generative artificial intelligence, or gen AI.

Whether your first encounter was with ChatGPT, Gemini or Copilot, many of us felt the same jolt of surprise. Here was a bot that seemed to understand us in a way we didn't expect. That has made Musk's dream of a robot companion feel if not close then certainly closer.

Imagine leafing through a catalogue of robots the way we browse for home appliances. If a personal robot still feels too expensive, perhaps we might hire one part time. Maybe a dance instructor that doubles as a therapist. Families could club together to buy a robot for an elderly relative. Some people might even buy one for themselves.

The future Musk describes isn't just mechanical, it's emotional.

Why the humanoid shape matters

The idea of robots that look like us can seem creepy and threatening. But there's also a practical explanation for the drive to make robots that look like us.

A dishwasher is essentially a robot but you have to load it yourself. A humanoid robot with hands and fingers could clear the table, load the dishwasher and then feed the pets too. In other words, engineers create humanoid robots because the world is designed for human bodies.

But the humanoid form also carries an emotional charge. A machine with a face and limbs hints at something more than functionality. It's a promise of intelligence, empathy or companionship. Optimus taps into that deep cultural imagery. It is part practical engineering, part theatre and part invitation to believe we are close to creating machines that can live alongside us.

There are moments when a personal robot might be genuinely welcoming. Anyone who has been ill, or cared for someone who is, can imagine the appeal of a helper that preserves dignity and independence. Robots, unlike humans, are not born to judge. But there is also a risk in outsourcing too much of our social world to machines.

If a robot is always there to tidy up the mess, practical or emotional, we may lose some of the tolerance and empathy that come from living among other people.

That is where the question of design becomes crucial. In the most dystopian version of life with generative AI-powered, chatty, dexterous robots, we retreat indoors, sealed into our homes and attended to by machines that are endlessly "understanding" and quietly adoring. Convenience is maximised, but something else is lost.

If sociability really does matter — if it is worth a little extra inconvenience to practise being human with other humans rather than only with chatbots — then the challenge becomes a practical one. How do we engineer a future that nudges us towards one another, instead of gently pulling us apart?

One option is to rethink where conversation lies. Rather than building all-purpose, ever-chatty assistants into every corner of our lives, we could distribute AI across devices and limit what those devices talk about. For example, a washing machine might discuss laundry, while a navigation system might discuss routes. But open-ended chatter, the kind that shapes identity, values and relationships, remains something that people do with people.

At a collective level, this kind of design choice could reshape workplaces and shared spaces, turning them back into environments that cultivate human conversation. That is, of course, only possible if people are encouraged to show up in person, and to put their phones away.

The real design challenge is not how to make machines more attentive to us, but how to make them better at guiding us back towards one another

So, it is worth asking what kind of domestic future we are quietly building. Will the robots we invite inside help us connect, or simply keep us company?

Good bots, bad bots

A good bot could help a socially anxious child get to school. It may nudge a lonely teenager towards local activities. Or it may tell a cantankerous old person: "There's a crime club starting in an hour at the library. We can pick up a paper on the way."

A bad bot leaves us exactly where we are: increasingly comfortable with a machine and less comfortable with each other.

Musk's humanoid dream may yet become real. The question is whether machines like Optimus will help us build stronger communities, or quietly erode the human connections we need most.

This edited article is republished from The Conversation under a Creative Commons license. Read the original article.

Engineers have built a detachable robotic hand that can crawl into hard-to-reach spaces to retrieve objects.

The spider-like device can be fitted with multiple fingers and is reversible, meaning it can grip objects in both directions and operate as though it has two functional palms.

In a study published Jan. 20 in the journal Nature Communications, scientists at the Swiss Federal Technology Institute of Lausanne (EPFL) said the robot was designed to combine two capabilities that are usually kept separate in robotics: manipulation and locomotion.

The hand can also operate without a full, mobile robot base. Instead, it simply detaches itself from a robotic arm and scuttles off to wherever it is needed, picks up the object and returns to reattach itself.

This could make it useful in situations where robots need to reach or retrieve objects from spaces that are too tight or too dangerous for human arms to access, such as industrial or exploratory environments and areas affected by disasters, the researchers noted.

"We can easily see the limitations of the human hand when attempting to reach objects underneath furniture or behind shelves, or performing simultaneous tasks like holding a bottle while picking up a chip," study co-author Aude Billard, head of the Learning Algorithms and Systems Laboratory in EPFL's School of Engineering, said in a statement.

"Likewise, accessing objects positioned behind the hand while keeping the grip stable can be extremely challenging, requiring awkward wrist contortions or body repositioning."

Handy robotics

Robotic appendages, like hands, have been built before. The human hand is, after all, considered a biological marvel; it's given humans the dexterity needed to make tools, prepare food and build shelter, all of which have proved key to our survival as a species.

Still, human hands do have some distinctly biological limitations, the scientists noted. For example, our asymmetrical thumbs and our hands' permanent attachment to our arms.

EPFL's robot is essentially a self-contained system that can either act like a normal "gripper" on the end of an arm or detach itself and scurry about on its own. The design draws inspiration from nature, with the researchers likening it to how an octopus uses its arms to move across the seafloor and open shells, or how a praying mantis uses its spiked forearms to both move around and catch prey.

In experiments, the researchers demonstrated the hand performing various standard gripping and grasping exercises, as well as accomplishing more fiddly feats of dexterity that humans routinely struggle with. These included holding multiple objects at once or gripping objects without using the thumb or forefinger.

Not only can the robotic hand grasp up to four objects simultaneously, it can also crawl across the floor while carrying items on its "back."

Each of the robot's fingers is driven by small electric motors and linked by lightweight 3D-printed joints, allowing them to curl and spread much like human fingers. Unlike a human hand, however, each finger joint can bend both forward and backward, allowing it to grab objects in both directions and "flip" its working orientation without needing to rotate at the wrist.

The fingertips are capped with a soft silicone layer to add friction, making it easier to hold objects securely and maintain traction when crawling. "There is no real limitation in the number of objects it can hold; if we need to hold more objects, we simply add more fingers," Billard said.

The hand reattaches itself to the arm using a "snap-and-lock" system. Magnets help align the connectors, and a small motor drives a locking bolt that secures the joints.

The system could eventually be adapted for human prosthetics or "extra limb" augmentation, the team said, though this isn't the focus of the current prototype.

"The symmetrical, reversible functionality is particularly valuable in scenarios where users could benefit from capabilities beyond normal human function," Billard said.

"For example, previous studies with users of additional robotic fingers demonstrate the brain's remarkable adaptability to integrate additional appendages, suggesting that our non-traditional configuration could even serve in specialized environments requiring augmented manipulation abilities."

Caltech engineers have developed a multimodal robot system — a humanoid robot with a transforming drone that launches off its back.

Sitting on the back of the humanoid robot, a Unitree G1 machine, the drone, called M4, can transform — switching between driving and flight modes. We’re not talking Optimus Prime or Megatron here; instead, the drone-expelling bot is more like Soundwave, a Decepticon that housed different, mini transformers, like drones, in its chest.

The system is the culmination of a three-year collaboration between Caltech's Center for Autonomous Systems and Technologies (CAST) and the Technology Innovation Institute (TII) in Abu Dhabi, United Arab Emirates.

The humanoid can walk (although we have seen smoother movers) and it can tackle stairs and navigate its way to wherever it has sent the drone, though at a much slower pace.

"Right now, robots can fly, robots can drive, and robots can walk. Those are all great in certain scenarios," Aaron Ames, director of CAST and a professor of aerospace and engineering at Caltech, said in a statement. "But how do we take those different locomotion modalities and put them together into a single package, so we can excel from the benefits of all these while mitigating the downfalls that each of them have?"

The challenge here lay in how the team brought different robots together so that they could become one system but still offer different functionalities. The drone was built by a CAST team led by Mory Gharib, while Ames and his lab configured the humanoid robot.

M4 is capable of reconfiguring its body (transforming) into several different types of motion — it can assess the environment it needs to enter and automatically select the most effective combinations of motion to maneuver. M4 can roll on four wheels, turn its wheels into rotors and fly, stand meerkat-style on two wheels, "walk" by using its wheels like feet, use two rotors to help it roll up steep slopes on two wheels, and also simply tumble towards its destination.

M4’s ability to repurpose its appendages as wheels, legs, or thrusters is a key feature. When M4 needs to fly, all four wheels fold up, and the propellers lift the robot off the ground — or off the back of the humanoid robot, which bends forward when deploying M4.

The overarching goal of the collaboration is to make such autonomous systems safer and more reliable. If we are going to have robots all around us, Ames suggested, more work needs to be focused on making them reliable.

"We're thinking about safety-critical control, making sure we can trust our systems, making sure they're secure," Ames said in the statement. "We have multiple projects that extend beyond this one that study all these different facets of autonomy, and these problems are really big. By having these different projects and facets of our collaboration, we are able to take on these much bigger problems and really move autonomy forward in a substantial and concerted way."

A robot chair revealed at the Japan Mobility Show 2025 can navigate complicated environments on its four articulated legs.

While the chair is still a prototype, it aims to allow users with limited mobility to climb stairs or cross other obstacles that would be impassable by traditional wheelchairs. It's also capable of lifting the user so they can access cars and other elevated vehicles or platforms.

Developed by Toyota, the Walk Me prototype features four foldable legs and a seat designed to support proper posture. The legs are swaddled in a soft, colorful material that serves the dual purpose of protecting the sensitive internals (like sensors and motors) from external damage, while also giving the unit a pleasant, approachable aesthetic.

The legs are wholly independent, with each bending, lifting or folding to aid manoeuvrability. When not in use, the legs can also fold away neatly beneath the robot, allowing it to be packed into a car or luggage for easy transport. The system can also unfold and stabilize itself without user assistance.

Described as an "autonomous wheelchair," the bot is packed with a number of features that allow it to navigate difficult terrain by mimicking the movement of four-legged animals like crabs. These include LiDAR systems that use laser light to measure distances and create highly accurate, detailed three-dimensional representations of objects and environments, which the robot utilizes to dodge obstacles or deal with uneven surfaces.

When climbing stairs, the unit first tests the height with its front legs before pushing upward with its rear limbs. There are also built-in collision radars to avoid contact with people or objects.

Additionally, the Walk Me has built-in weight sensors to ensure that the user remains in a stable, seated position. Toyota's engineers studied the way people naturally navigate stairs and how they distribute their weight when moving around or over obstacles. If the robot senses an imbalance, it can adjust both its legs as well as the tilt of the seat itself to ensure the user is comfortable and secure.

There are also a number of manual control options. Handles are attached to the seat that allow the user to guide the robot's direction. Alternatively, a digital interface provides specific buttons to control locomotion precisely. The Walk Me will also respond to voice commands that include preset destinations like "living room" and speed controls like "slower" or "faster."

The unit is powered by a battery concealed behind the seat, which can power it for an entire day of operation. The battery is charged by plugging it into a standard wall outlet overnight.

The Walk Me was part of a broader product lineup shown by Toyota at the Tokyo Mobility Show, which also included an autonomous, self-driving car for kids and a "Land Cruiser of wheelchairs" with extra-rugged, all-terrain tires and a durable frame. According to Top Gear, the wheelchair was inspired by Toyota's chairman Akio Toyoda who, at 69, wants to be able to "drift, do donuts and race off-road into his retirement."

Chinese electric vehicle (EV) maker Xpeng has unveiled a new humanoid robot with such lifelike movements that company representatives felt compelled to slice it open onstage to prove a human wasn't hiding inside.

Fortunately for the audience, there wasn't. Instead, the robot, named "IRON," features a flexible, humanlike spine, articulated joints and artificial muscles that allow it to move with a model-like swagger.

This is thanks to Xpeng's custom artificial intelligence (AI) robotics architecture, which enables it to interpret visual inputs and respond physically without needing to first translate what it sees into language.

Speaking during IRON's unveiling at Xpeng's AI Day in Guangzhou on Nov. 5, China, He Xiaopeng, chairman and CEO of Xpeng Motors, suggested that IRON's appearance was designed to be recognizably human — if slightly unsettling.

The machine is equipped with 82 degrees of freedom, including 22 in each hand, allowing it to bend, pivot and gesture at multiple points throughout its body, representatives said in a statement.

It's powered by three custom AI chips that give it a combined 2,250 trillion operations per second (TOPS) of computing power, which Xpeng says makes it one of the most powerful humanoid robots developed to date. For comparison, Intel's Core Ultra 200V series processor, fitted into some of the best laptops, can achieve just 120 TOPS.

IRON man

IRON is based on what its creators call a "born from within" design, a concept that reflects the robot’s design mimicking the human body from the inside out.

The robot features an internal endoskeleton and bionic muscle structure capable of supporting different body types, ranging from slim to stocky, which users can customize. Its outer layer is also made from "full-coverage" synthetic skin, He said during the presentation, making the robot "feel warmer and more intimate."

"The next generation has very flexible bones, solid bionic muscles, and soft skin. We hope it can have a similar height and proportions to human beings," He said. "In the future, robots will be life partners and colleagues. I suspect that, just like when you buy a car, you can choose different colors, exteriors, and interiors. In the future, when you buy a robot, you can choose the sex, hair length, or clothing for your desired purpose."

According to Xpeng, IRON is also the first humanoid robot in the world to run on an all-solid-state battery. Solid-state batteries use ceramics or polymers instead of the flammable liquids in conventional lithium-ion batteries, making them safer for the enclosed environments where the robot is designed to operate.

IRON is destined for mass production, although Xpeng ruled out household chores for the immediate future, pointing out that a humanoid robot operating in messy or unpredictable households could pose safety risks. Instead, it will debut in commercial settings such as stores, offices and company showrooms, with the first models expected to appear in Xpeng locations in 2026.

The announcement forms part of Xpeng's broader push into "physical AI," which aims to bring together robotics, autonomous vehicles and AI development under a unified platform. Earlier this year, the company revealed a prototype flying car designed to launch from a Cybertruck-style mobile base.

Humanoid robots have been having something of a moment in recent months. In October, Chinese robotics startup Unitree debuted its pirouetting, karate-kicking H2 model. Unlike IRON, Unitree's bot has yet to be given an official release date, meaning Xpeng's bot may well beat it to the shop floor (or office reception).

Researchers in South Korea have built an artificial muscle that can lift approximately 4,000 times its own weight. They say it can be used in future humanoid robots.

A key breakthrough with the muscle's design is its ability to be flexible or taut when needed, which is a first for this field of research. The scientists outlined their findings in a study published Sept. 7 in the journal Advanced Functional Materials.

"This research overcomes the fundamental limitation where traditional artificial muscles are either highly stretchable but weak or strong but stiff," lead study author Hoon Eui Jeong, a professor of mechanical engineering at the Ulsan National Institute of Science and Technology (UNIST), said in a statement. "Our composite material can do both, opening the door to more versatile soft robots, wearable devices, and intuitive human-machine interfaces."

Artificial muscles are often limited by an inability to be flexible or taut; they need to be stretchable while still offering enough energy output, or else their work densities are limited. But soft artificial muscles are believed to be transformative because they're lightweight, mechanically compliant, and capable of multidirectional actuation (movement).

When the researchers say "work density," they refer to how much energy per unit volume the muscle can deliver. Achieving high values alongside high stretchability is where the challenge lies for artificial muscles.

Do you even lift, robo?

The scientists described their artificial muscle as a "high-performance magnetic composite actuator," which means it's a complex chemical combination of polymers that link together to mimic the pull and release of muscles.

One of these polymers can have its level of stiffness altered and sits in a matrix that has magnetic microparticles on the surface that can also be controlled. This enables the muscle to be animated and controlled through the tunable stiffness, thus allowing it to be moved.

The researchers' new design integrates two distinct cross-linking mechanisms. The first is a covalently bonded chemical network (two or more atoms that share electrons to achieve a more stable configuration) and a reversible, physically interacting network. The two mechanisms, developed in this way, provide the durability for the muscle to work long-term, the researchers said in the study.

The trade-off between stiffness and stretchability is effectively solved by a dual cross-linking architecture, and the physical network is further reinforced by incorporating a type of microparticle (NdFeB) on the surface of the muscle that can be given a function via a colorless liquid (octadecyltrichlorosilane). The particles are dispersed throughout the polymer matrix.

The composite muscle becomes stiff when bearing heavy loads and softens when it needs to contract. In its stiffened state, the artificial muscle, which weighs just 0.04 ounce (1.13 grams), can support up to 11 pounds (5 kilograms) — roughly 4,400 times its own weight.

A human muscle contracts at approximately 40% strain, but the synthetic muscle achieves a strain of 86.4% — over double that of the human muscle, the researchers said in the study. This enables a work density of 1,150 kilojoules per meter cubed — 30 times higher than human tissue is capable of.

The researchers used a uniaxial tensile test to measure the strength of their artificial muscle. A type of mechanical test that applies a pulling force to a subject until it fractures – the elongation is measured against the applied force to find its ultimate tensile strength.

A squishy robotic "eye" can focus automatically in response to light, without any external power. The ultrapowerful robotic lens is sensitive enough to distinguish hairs on an ant's leg or the lobes of a pollen grain.

The lens could usher in "soft" robots with powerful vision that would not need electronics or batteries to operate. Soft robotics can be used in a wide range of different applications, from wearable technology that can integrate with the human body to autonomous devices that can operate in uneven terrain or hazardous spaces, said study first author Corey Zheng, a doctoral student in biomedical engineering at the Georgia Institute of Technology. Traditional, electrically powered robots use rigid sensors and electronics to see the world.

But "if you're looking at robots that are softer, they're squishy, they maybe don't use electricity, then you have to think about how you're going to do sensing with these robots," Zheng told Live Science.

The lens is made of a hydrogel, which contains a framework of polymers that can trap and release water, allowing the hydrogel to move between more liquid-like and more solid-like states. In this case, the hydrogel responds to heat by releasing water and shrinking when it's warmed up, and absorbing water and swelling when it's cooled.

The researchers fabricated a ring of hydrogel around a silicon polymer lens, situating the eye-like design in a larger frame. The mechanical structure is similar to the configuration of the human eye, Zheng said.

The hydrogel is embedded with tiny particles of graphene oxide, which are dark-colored and absorb light. When light of intensity equivalent to sunlight hits the graphene oxide, the particles heat up and warm the hydrogel, which shrinks and stretches, pulling the lens to focus it. When the light source is removed, the hydrogel swells and releases the tension on the lens. The hydrogel reacts to light across the visible spectrum.

In a new paper published today (Oct. 22) in the journal Science Robotics, Zheng and his doctoral adviser Shu Jia, a biomedical engineer at Georgia Tech, found that this lens could be used instead of the glass lens in a traditional light microscope to distinguish tiny details. For instance, the lens could image the 4-micrometer gap between a tick's claws, see 5-micrometer filaments of fungus, and detect the 9-micrometer stubble on an ant's leg.

More exciting, Zheng said, is that the researchers are now integrating the lens into a microfluidic system of valves made from the same responsive hydrogel. That means the light used to make the image can also serve to power an intelligent, autonomous camera system, Zheng said.

And because the hydrogel is adaptable, the lens may be able to "see" well beyond what the human eye can detect. For instance, it might be able to mimic the ability of a cat's vertical eye to detect camouflaged objects, or to copy a cuttlefish's odd W-shaped retina, which enables it to see colors humans can't.

"We can actually control the lens in really unique ways," Zheng said.

Chinese engineers have published a video of a robot that moves so fluidly you'd be forgiven for thinking it's artificial intelligence (AI) generated.

Yet it’s very much real. The video — which is quickly amassing views on social media — actually shows off the latest creation by China's Unitree Robotics. Named Unitree H2 "Destiny Awakening", the smooth-moving humanoid robot stands 5 foot 11 inches (180 centimeters) tall and weighs 154 pounds (70 kilograms) — making it around the size and weight of an adult man.

In the video, the bipedal H2 model is seen posing, pirouetting and pulling off some remarkably deft karate moves, before strutting its stuff on an improvised catwalk. The smooth, silver-grey model bears a striking resemblance to Sonny, the AI-driven tritagonist in the 2004 sci-fi thriller, I, Robot — a likeness that has not gone unnoticed by internet users.

Specifics on the H2 remain thin on the ground, but in the captions accompanying the YouTube video, Unitree states that its new "bionic humanoid" is "born to serve everyone safely and friendly."

In a disclaimer at the end of the video, the company also notes that "features vary by models and versions" and requests that "all users refrain from making any dangerous modifications or using the robot in a hazardous manner." On that basis, we can safely assume that the H2 will be made available to purchase at some point in the future, though at the moment, a specific release date is lacking.

H2 Destiny follows the launch of Unitree's G1 robot earlier in August 2024. A video released in 2025 showed it pulling off a complicated side flip.

The G1 "Evolution V3.0" stands at just over 4 foot 3 inches (130 cm) tall and weighs 77 pounds (35 kg). As well as sporting a 3D light detecting and ranging (lidar) and depth camera, G1 can run speeds of up to 4.5 mph (7.2 km/h) and incorporates 23 degrees of freedom — a measure of the number of individual movements the robot can make through its joints or axes.

Meanwhile, the H1 model — H2's direct predecessor — was released in 2023, claiming to be the world's fastest humanoid robot. A video published by Unitree at the time shows the robot running at a top speed of 7.4 miles per hour (11.9 km/h), though the startup claims the H1 can reach speeds of 11 mph (17.7 km/h).

Given the continuity in body size and design, the H2 is likely to build on many of the traits of its predecessor, which includes 360-degree depth sensing, 3D lidar and a front-mounted depth camera for navigation, along with 27 degrees of freedom.

Unitree also lists a maximum torque of 120 newton-meters in the arm joints and 360 newton-meters in the legs, giving the robot its fluid movements and eerily precise control over its limbs. This design enables the robot to perform tasks like lifting and moving objects, accessing hard-to-reach areas and traversing difficult terrain.

Google DeepMind has unveiled a pair of artificial intelligence (AI) models that will enable robots to perform complex general tasks and reason in a way that was previously impossible.

Earlier this year, the company revealed the first iteration of Gemini Robotics, an AI model based on its Gemini large language model (LLM) — but specialized for robotics. This allowed machines to reason and perform simple tasks in physical spaces.

The new models, dubbed Gemini Robotics 1.5 and Gemini Robotics-ER 1.5, greatly expand on the capabilities of the original version to handle multistep, "long-horizon" tasks and are a significant milestone towards robots assisting people in real-world use cases.

The baseline example Google points to is the banana test. The original AI model was capable of receiving a simple instruction like "place this banana in the basket," and guiding a robotic arm to complete that command.

Powered by the two new models, a robot can now take a selection of fruit and sort them into individual containers based on color. In one demonstration, a pair of robotic arms (the company's Aloha 2 robot) accurately sorts a banana, an apple and a lime onto three plates of the appropriate color. Further, the robot explains in natural language what it's doing and why as it performs the task.

"We enable it to think," said Jie Tan, a senior staff research scientist at DeepMind, in the video. "It can perceive the environment, think step-by-step and then finish this multistep task. Although this example seems very simple, the idea behind it is really powerful. The same model is going to power more sophisticated humanoid robots to do more complicated daily tasks."

AI-powered robotics of tomorrow

While the demonstration may seem simple on the surface, it demonstrates a number of sophisticated capabilities. The robot can spatially locate the fruit and the plates, identify the fruit and the color of all of the objects, match the fruit to the plates according to shared characteristics and provide a natural language output describing its reasoning.

It's all possible because of the way the newest iterations of the AI models interact. They work together in much the same way a supervisor and worker do.

Google Robotics-ER 1.5 (the "brain") is a vision-language model (VLM) that gathers information about a space and the objects located within it, processes natural language commands and can utilize advanced reasoning and tools to send instructions to Google Robotics 1.5 (the "hands and eyes"), a vision-language-action (VLA) model. Google Robotics 1.5 matches those instructions to its visual understanding of a space and builds a plan before executing them, providing feedback about its processes and reasoning throughout.

The two models are more capable than previous versions and can use tools like Google Search to complete tasks.

The team demonstrated this capacity by having a researcher ask Aloha to use recycling rules based on her location to sort some objects into compost, recycling and trash bins. The robot recognized that the user was located in San Francisco and found recycling rules on the internet to help it accurately sort trash into the appropriate receptacles.

Another advance represented in the new models is the ability to learn (and apply that learning) across multiple robotics systems. DeepMind representatives said in a statement that any learning gleaned across its Aloha 2 robot (the pair of robotics arms), Apollo humanoid robot and bi-arm Franka robot can be applied to any other system due to the generalized way the models learn and evolve.

"General-purpose robots need a deep understanding of the physical world, advanced reasoning, and general and dexterous control," the Gemini Robotics Team said in a technical report on the new models. That kind of generalized reasoning means that the models can approach a problem with a broad understanding of physical spaces and interactions and problem-solve accordingly, breaking tasks down into small, individual steps that can be easily executed. This contrasts with earlier approaches, which relied on specialized knowledge that only applied to very specific, narrow situations and individual robots.

The scientists provided an additional example of how robots could help in a real-world scenario. They presented an Apollo robot with two bins and asked it to sort clothes by color — with whites going into one bin and other colors into the other. They then added an additional hurdle as the task progressed by moving the clothes and bins around, forcing the robot to reevaluate the physical space and react accordingly, which it managed successfully.

A Chinese company has released footage of a lifelike robot head that blinks, nods, and looks around in a convincing imitation of a real human face.

The video, posted to YouTube, shows the robot head’s quizzical expression as it appears to take in its surroundings. The head, created by robotics company AheadForm, might find use in research on human-robot interactions or in various industries, including customer service and entertainment.

Founded in 2024, AheadForm aims to make interactions between humans and robots feel more natural and engaging, according to the company’s website. The company aims to integrate artificial intelligence (AI) systems such as large language models (LLMs) with realistic robotic heads so that robots can understand and respond to people in real-time.

"Our current focus is on creating sophisticated humanoid robot heads that can express emotions, perceive their environment, and interact seamlessly with humans," company representatives said on the website.

AheadForm has produced multiple series of robot forms, including its "Elf" line of pointy-eared figures with precise control systems and a more human-looking "Lan Series" of bots designed to prioritize cost-efficiency and intuitive movements.

In building a robotic head that can integrate with AI, the company aims to make robots more approachable and relatable. That could be useful for industries such as customer service, education, and healthcare, "where building trust and rapport with human users is essential," according to the website.

Researchers from AheadForm are already working on how this might play out. In a 2024 study published in the journal Science Robotics , AheadForm founder Yuhang Hu and colleagues designed a robot that could analyze, predict, and mimic human facial expressions in real-time.

To achieve the subtle, precise facial movements in the video, AheadForm uses brushless motors. These tiny devices run quietly and work together to create the lifelike twitches and glances that make the head seem realistic.

Inside this particular model, the Origin M1, there are up to 25 tiny motors that control the head’s expressions. The bot also has cameras embedded in its pupils to help it "see" its environment, along with built-in speakers and microphones it can use to interact with users on the fly.

The robot heads are not yet commercially available.

Scientists have trained a four-legged robot to play badminton against a human opponent, and it scuttles across the court to play rallies of up to 10 shots.

By combining whole-body movements with visual perception, the robot, called "ANYmal," learned to adapt the way it moved to reach the shuttlecock and successfully return it over the net, thanks to artificial intelligence (AI).

This shows that four-legged robots can be built as opponents in "complex and dynamic sports scenarios," the researchers wrote in a study published May 28 in the journal Science Robotics.

ANYmal is a four-legged, dog-like robot that weighs 110 pounds (50 kilograms) and stands about 1.5 feet (0.5 meters) tall. Having four legs allows ANYmal and similar quadruped robots to travel across challenging terrain and move up and down obstacles.

Researchers have previously added arms to these dog-like machines and taught them how to fetch particular objects or open doors by grabbing the handle. But coordinating limb control and visual perception in a dynamic environment remains a challenge in robotics.

Related: Watch a 'robot dog' scramble through a basic parkour course with the help of AI

"Sports is a good application for this kind of research because you can gradually increase the competitiveness or difficulty," study co-author Yuntao Ma, a robotics researcher previously at ETH Zürich and now with the startup Light Robotics, told Live Science.

Teaching a new dog new tricks

In this research, Ma and his team attached a dynamic arm holding a badminton racket at a 45-degree angle onto the standard ANYmal robot.

With the addition of the arm, the robot stood 5 feet, 3 inches (1.6 m) tall and had 18 joints: three on each of the four legs, and six on the arm. The researchers designed a complex built-in system that controlled the arm and leg movements.

The team also added a stereo camera, which had two lenses stacked on top of each other, just to the right of center on the front of the robot's body. The two lenses allowed it to process visual information about the incoming shuttlecocks in real time and work out where they were heading.

The robot was then taught to become a badminton player through reinforcement learning. With this type of machine learning, the robot explored its environment and used trial and error to learn to spot and track the shuttlecock, navigate toward it and swing the racket.

To do this, the researchers first created a simulated environment consisting of a badminton court, with the robot's virtual counterpart standing in the center. Virtual shuttlecocks were served from near the center of the opponent's half of the court, and the robot was tasked with tracking its position and estimating its flight trajectory.

Then, the researchers created a strict training regimen to teach ANYmal how to strike the shuttlecocks, with a virtual coach rewarding the robot for a variety of characteristics, including the position of the racket, the angle of the racket's head, and the speed of the swing. Importantly, the swing rewards were time-based to incentivize accurate and timely hits.

The shuttlecock could land anywhere across the court, so the robot was also rewarded if it moved efficiently across the court and if it didn't speed up unnecessarily. ANYmal's goal was to maximize how much it was rewarded across all of the trials.

Based on 50 million trials of this simulation training, the researchers created a neural network that could control the movement of all 18 joints to travel toward and hit the shuttlecock.

A fast learner

After the simulations, the scientists transferred the neural network into the robot, and ANYmal was put through its paces in the real world.

Here, the robot was trained to find and track a bright-orange shuttlecock served by another machine, which enabled the researchers to control the speed, angles and landing locations of the shuttlecocks. ANYmal had to scuttle across the court to hit the shuttlecock at a speed that would return it over the net and to the center of the court.

The researchers found that, following extensive training, the robot could track shuttlecocks and accurately return them with swing speeds of up to approximately 39 feet per second (12 meters per second) — roughly half the swing speed of an average human amateur badminton player, the researchers noted.

ANYmal also adjusted its movement patterns based on how far it had to travel to the shuttlecock and how long it had to reach it. The robot did not need to travel when the shuttlecock was due to land only a couple of feet (half a meter) away, but at about 5 feet (1.5 m), ANYmal scrambled to reach the shuttlecock by moving all four legs. At about 7 feet (2.2 m) away, the robot galloped over to the shuttlecock, producing a period of elevation that extended the arm's reach by 3 feet (1 m) in the direction of the target.

"Controlling the robot to look at the shuttleclock is not so trivial," Ma said. If the robot is looking at the shuttlecock, it can't move very fast. But if it doesn't look, it won't know where it needs to go. "This trade-off has to happen in a somewhat intelligent way," he said.

Ma was surprised by how well the robot figured out how to move all 18 joints in a coordinated way. It's a particularly challenging task because the motor at each joint learns independently, but the final movement requires them to work in tandem.

The team also found that the robot spontaneously started to move back to the center of the court after each hit, akin to how human players prepare for incoming shuttlecocks.

However, the researchers noted that the robot did not consider the opponent's movements, which is an important way human players predict shuttlecock trajectories. Including human pose estimates would help to improve ANYmal's performance, the team said in the study. They could also add a neck joint to allow the robot to monitor the shuttlecock for more time, Ma noted.

He thinks this research will ultimately have applications beyond sports. For example, it could support debris removal during disaster relief efforts, he said, as the robot would be able to balance the dynamic visual perception with agile motion.

Humanoid robots can pick up objects, run, and even play sports (to a limited degree), but can they master the rhythm of a drum kit? Thanks to new research, they sort of can.

Robot Drummer is a humanoid robot avatar capable of “expressive, high-precision drumming”, scientists at the Politecnico di Milano said in a new study uploaded July 15 to the preprint ArXiv database.

This was a concept that was spontaneously conceived over coffee by co-authors Asad Ali Shahid and Loris Roveda, they told Tech Explore.

"We were discussing how humanoid robots have become increasingly capable at a wide range of tasks, but rarely engage in creative and expressive domains," Ali Shahid said. "That raised a fascinating question: what if a humanoid robot could take on a creative role, like performing music? Drumming seemed like a perfect frontier, as it's rhythmic, physical, and requires rapid coordination across limbs."

Robot Drummer isn’t a robot itself; rather, it's a simulation that uses the G1 Unitree robot as a model to build a system that a future humanoid robot could eventually use. The 3-D visualised G1 robot plays a color-coded drum kit.

In a video published on YouTube, Robot Drummer is shown mimicking the drum beats to two well-known songs. There's something a little off about the sound — perhaps to make the drumming more accessible to the ear. Its rendition of “In the end”, by Linkin Park sounds just off the rhythm, while its attempt at "Roxanne" by the Police sounds more matched to the rest of the arrangement.

Related: Robots awkwardly race, fight and flop around in China's first World Humanoid Robot Games

Learning to play any musical instrument, particularly the drums, presents a unique challenge, the researchers said in the study. It is a multifaceted task that needs split-second timing, rapid contacts, and the combined coordination of hands and feet — all of which needs to be sustained for several minutes.

Trial and error was the basis for Robot Drummer’s learning, with the researchers employing reinforcement learning (RL), a type of artificial intelligence (AI) machine learning process in which an agent learns to make decisions by interacting with its environment.

The agent is rewarded or punished for its actions with a mathematical signal, and it’s often a positive number that signifies a correct action. The aim is to earn more rewards by finding the optimal policy. For Robot Drummer, each song was represented as a chain of precisely timed “contact events,” according to the researchers, which informed the robot what drum strikes were needed and where.

Learned behaviours, the researchers noted, included specific drumming strategies, such as cross-arm strikes and adaptive stick assignments (using a drum stick for a particular sound). This, the researchers said, demonstrated the potential of reinforcement learning for robots in creative domains such as music.

The researchers said that Robot Drummer was put through extensive experiments with over 30 popular songs across pop, rock, metal, and jazz.

A viral story spread through the tabloids this week, proclaiming that a CEO of a technology company in China was in the midst of developing a "pregnancy robot." A prototype of the bot would be ready by next year, the CEO projected, and would be designed to sustain a human pregnancy from conception to birth, around 10 months.

The robot and the related story were — as you might have guessed — pure fiction. But the viral tale nonetheless raised questions about whether designing such a robot could be possible with today's technology, and what ethical quandaries that possibility might raise, in turn.

If a "pregnancy robot" existed, would you consider using one? Take our poll below, and let us know what you think of this sci-fi-sounding idea in the comments below. If you chose "maybe," what would sway your decision?

Related: 'Pregnancy robot from China' is fake, but is the technology behind it possible?

Related stories

—'I would never let a robot incubate my child': Poll on 'pregnancy robots' divides Live Science readers

—The gut 'remodels' itself during pregnancy, study finds

—Scientists reveal surprising factor that may prolong pregnancy

Bizarre imagery flooded newsfeeds this week as word spread about a new "pregnancy robot" in development in China. The images featured plump human babies curled inside the bellies of chrome-plated robots complete with visible wiring and ample curves, despite their lack of mammary glands.

Many outlets — including Newsweek, The Economic Times, and ChosunBiz — named a Chinese outlet, Kuai Ke Zhi, as the story's source. Zhang Qifeng, the developer of the bot intended to carry a pregnancy from conception to birth, reportedly told the outlet that a prototype would be ready as early as 2026 at a price point under 100,000 yuan (around $13,900 USD).

"Some people don't want to get married but still want a 'wife'; some don't want to be pregnant but still want a child," Zhang said, according to Newsweek. "Mature" artificial womb technology just "needs to be implanted in the robot's abdomen so that a real person and the robot can interact to achieve pregnancy," he said, according to ChosunBiz. The nature of that human-robot interaction was not detailed.

Depending on the article, Zhang was cited as the CEO or founder of Kaiwa Technology, a Guangzhou-based company, or as a PhD affiliated with Singapore's Nanyang Technological University (NTU). Finding no online evidence of Kaiwa Technology, Live Science contacted NTU about their purported affiliation with Zhang Qifeng.

"We would like to inform you that no one by the name of 'Zhang Qifeng' graduated from NTU with a PhD," an NTU spokesperson told Live Science via email. "Our checks also showed no such 'gestation robot' research has been conducted at NTU."

As you may have suspected — and as Snopes has also confirmed — the pregnancy robot isn't real. But the viral story raised questions about the potential of artificial-womb technology. Would it be possible to build a pregnancy robot? Or is the concept pure science fiction? Live Science spoke with experts about the idea, discussing whether it would technically be possible and whether anyone should even try.

"Should we do it? My answer would be categorically 'no,'" said Dr. Harvey Kliman, director of the Reproductive and Placental Research Unit at Yale University School of Medicine. "That being said, intellectually, I think it's interesting to think about the challenges because it helps us actually reflect about what is the beauty and miracle of a normal pregnancy."

Related: If 'pregnancy robots' were real, would you use one?

Artificial wombs?

The pregnancy robot is fake, but scientists have been developing artificial wombs. At Children's Hospital of Philadelphia (CHOP), scientists are developing a womb-like device called the "extra-uterine environment for newborn development," or EXTEND. The eventual hope is to support babies who are born extremely premature, between 23 and 28 weeks of gestation.

Recent advances have decreased the death rate associated with preterm birth, but health issues, including chronic lung disease and neurodevelopmental problems, remain a big concern for babies born that early. To decrease these risks, CHOP researchers aim to create a uterus-like environment that babies can be placed in after delivery to help them over the 28-week mark.

The device includes a bag full of amniotic fluid, which is made in the lab and contains key nutrients and growth factors. The umbilical cord is attached to an "external oxygenator" that partly stands in for the placenta, facilitating an exchange of oxygen and carbon dioxide. Within the device, a baby would be insulated from changes in temperature, pressure and light, as well as from exposure to germs.

EXTEND has so far been tested with lambs. In a 2017 paper, the team showed that fetal lambs could be supported in the device for a month and that their development continued much as it would have in the womb. In a 2024 paper, they collaborated with Duke University researchers to see how EXTEND impacted gene activity in the brain. The device helped preserve gene activity in premature lambs' brains so that it resembled that of lambs that remained in the womb for much longer.

Meanwhile, some researchers are working on artificial placentas that would fulfill the same purpose as EXTEND, supporting premature babies. These devices have also been tested in lambs but are farther out from human trials than EXTEND is.

The EXTEND team aims to move into human trials soon, though there are questions about when and how it would be ethical to test the device, given many premature babies have a fighting chance with existing technologies.

"In a human setting, a couple weeks is not enough," said Dr. Lusine Aghajanova, a fertility specialist and clinical associate professor of obstetrics and gynecology at Stanford Medicine, in reference to the length of the tests done so far in lambs. "What that study showed is that it's a concept that can be possible, but it's more complicated than we think," she told Live Science.

Throughout their efforts, the EXTEND developers have stressed that the device is intended as a bridge for premature babies moving from the womb into the world; it is not intended to be used to push fetal viability earlier than 23 weeks. Of course, there's a big difference between keeping an ailing baby alive and gestating a baby from conception, as the fake Kaiwa pregnancy robot was supposed to do.

Related: Should we rethink our legal definition of a human embryo?

Like a 'tomato plant suspended in water'

Presumably, a gestation robot might need to deal with a particularly tricky part of pregnancy: implantation.

In an unassisted pregnancy, an egg travels through a fallopian tube, where it is typically fertilized before moving into the uterus and implanting in the uterine wall. In fertility treatments, such as in vitro fertilization (IVF), a fertilized egg is placed into the uterus, where ideally, it then implants itself. That aspect of the treatment is not directly orchestrated by doctors.

In a robotic incubator, though, implantation may not be a critical step, Kliman said. "I'm thinking about hydroponics," he said. "Think of it as a tomato plant suspended in this bucket of water, right?"

Similar to the EXTEND studies with lambs, in theory, a human embryo could be suspended in fluid rather than embedded in something akin to the uterine wall, he said. The real challenge would be ensuring that the embryo remains suspended, so it can grow unimpeded, and that it's supplied with adequate nutrients and factors to grow. The placenta "doesn't actually have to attach to anything or burrow into anything" in that kind of setup, Kliman said.

Aghajanova, on the other hand, thinks recreating implantation would be key. "Implantation is absolutely important," she said. "It's the seed and the soil," referencing the embryo and the uterine lining. Abnormalities in the uterine lining can undermine both fetal growth and placental development, she argued, so some stand-in for the tissue would likely be necessary in a pregnancy robot.

Even if the implantation issue was addressed, delivering the right nutrients to the fetus at the right time could still be tricky.

Early in pregnancy — roughly up to the eighth week or so — glands in the uterine lining produce a nutritious "milk" for the developing embryo and placenta. At that point, blood flow from mother to womb is not fully established, in part, because the maternal blood is too oxygenated, Kliman said. "That high-oxygen state causes too many free radicals and would destroy the DNA of the dividing embryo," he said, so instead, the early womb maintains a low-oxygen state.

In a robot pregnancy, you'd need to carefully recreate that transition from low- to high-oxygen, as well as maintain the womb-like environment and keep it flush with nutrients for months on end. Maternal metabolism shifts dramatically in pregnancy, Aghajanova said, so it would be difficult to know what dose of oxygen is needed at any given stage of development.

"That's mind boggling to me. I'm just trying to think of any system that would work flawlessly for nine months and not get infected," meaning exposed to germs that could derail the whole process, Kliman said. "We're talking extreme challenges of just the machinery, the reliability, the nutrients, the getting rid of the waste — and again, I think the biggest problem, to be honest with you, would be infection."

While those aspects of an artificial womb would be complex, it might be simpler than human reproduction in other respects, Kliman mused. In human pregnancy, the embryo and placenta must be protected from the maternal immune system, lest they be targeted as foreign invaders. Additionally the uterus undergoes changes to create a suitable environment for the embryo, and the placenta releases hormones that prep the mammary glands to make milk. These factors would be irrelevant in a machine, he said.

There are questions about how the fetal immune system might develop differently in a robot. In a human pregnancy, antibodies pass from maternal blood to the fetus via the umbilical cord. That's why a variety of vaccines are given in late pregnancy: The vaccines stimulate maternal antibody production, those antibodies pass to the fetus, and newborns enter the world with some protection against dangerous infections. And additional, non-vaccine-induced antibodies also cross over.

Given robots lack immune systems, developers of a pregnancy robot might need to figure out how to replicate this process, perhaps with lab-made antibodies or with donated blood.

"I don't think that would be that hard," Kliman said. Some of those antibodies could be supplied after birth via breastmilk or through formula containing lab-made antibodies, he suggested. And you might take extra precautions to keep the baby from being exposed to too many germs too soon, if you knew that might be an issue, he added.

In the womb, the maternal immune system must avoid attacking the growing fetus while also protecting it from pathogens, Aghajanova said. The former issue might not be relevant in a robotic pregnancy, but that latter protection would still be necessary, she said.

Related: 'We know what to do; we just have to implement it.': Pregnancy is deadlier in the US than in other wealthy countries. But we could fix that.

Unanswered questions

An additional factor to consider might be the vaginal microbiome, which contains bacteria, fungi and other microorganisms that studies suggest impact the health of the developing fetus. Figuring out how and what to deliver into the robot to mimic this complex microbial community could be tricky, as we don't fully understand which species are helpful to fetal development.

With the loss of a microbiome would likely come the addition of plastic components within the robot itself, Aghajanova added. It's unclear how all that plastic could sway early development, but it likely wouldn't be great, she suggested.

Other quandaries cloud the notion of a pregnancy robot: How would the eggs and sperm be sourced? Whose gametes would be used to test and optimize the device, proving it could lead to a live birth? Where could such research be conducted, legally? Would fertilization occur inside the bot itself, or via a lab procedure akin to IVF? Who is tasked with maintaining and monitoring the bot's functions throughout the "pregnancy"? What would the birthing process look like for a robot, and would engineers be needed in the delivery room? Are there yet-unknown aspects of human pregnancy that a baby grown in a machine would miss out on?

In our current reality, many of these questions remain unanswerable — but they would certainly make a compelling premise for a sci-fi novel.

"It's good science fiction, but in a word, it doesn't exist yet," Aghajanova said. And "we are not that desperate [for such technology], especially in the U.S." In the United States, patients have access to fertility treatments, donor eggs and surrogates, she said.

For Kliman's part, he said that his main takeaway from this thought experiment is simply "what a miracle a pregnancy is."

"We're just little feeble beings trying to conceive of some way to do this artificially, and look at what nature has done," he said. "And it happens four million times a year in our country; we get a normal delivery. So that's the miracle."

This article is for informational purposes only and is not meant to offer medical advice.

Humanoid robots are racing, fighting and falling over in a first-of-its-kind World Humanoid Robot Games event held in China.

The Olympics-style competition features more than 500 robots from 16 different countries going head-to-head in sports such as running, soccer and kickboxing. The event also features more niche competitions, including medicine sorting and handling materials for cleaning services, Reuters reported.

An opening ceremony officially kicked off the games in Beijing on Thursday (Aug. 14), featuring robots dancing and playing musical instruments alongside human operators and companions. Robot athletes will now compete until the games come to a close on Sunday (Aug. 17).

Faced with an ageing population and stiff U.S. tech competition, China is investing billions of dollars into robotics. The games are a testament to the strides engineers are making in the field. However, spectators have also seen their fair share of robots moving awkwardly and falling over.

Human biology is very complicated, so building machines that can walk like us — let alone run and play sports — is difficult. For example, in robot soccer, participants didn't pass the ball to each other with Messi-like precision, but rather walked into the ball to clumsily knock it forward, occasionally stumbling over each other and having to be dragged off the pitch.

The robots are also slower than humans. The fastest robot to have ever run 1,500 meters, for example, finished in 6 minutes and 34 seconds, which is almost twice as long as the human record, which stands at 3 minutes and 26 seconds, according to France 24.

Related: China launches world's first robot that can run by itself 24/7 — watch it change its own batteries in unsettling new footage

We've already seen humanoid robots compete in sporting events. For example, in June, China hosted what was billed as the world's first humanoid robot combat competition, which saw kickboxing robots awkwardly knock seven bells out of each other. Robot aficionados might also be familiar with robots playing soccer and robots running half-marathons.

The new robot games bring together all of these sports and many more for the first time. They also provide engineers with an opportunity to test out their latest tech.

"You can test a lot of interesting new and exciting approaches in this contest," Max Polter, a member of the HTWK Robots football team from Germany, told Reuters. "If we try something and it doesn't work, we lose the game. That's sad but it is better than investing a lot of money into a product which failed."

Scientists have created a prototype robot that can grow, heal and improve itself by integrating material from its environment or by "consuming" other robots. It's a big step forward in developing robot autonomy, the researchers say.

The researchers coined the term "robot metabolism" to describe the process that enables machinery to absorb and reuse parts from its surroundings. The scientists published their work July 16 in the journal Science Advances.

"True autonomy means robots must not only think for themselves but also physically sustain themselves," study lead author Philippe Martin Wyder, professor of engineering at Columbia University, said in a statement.

"Just as biological life absorbs and integrates resources, these robots grow, adapt, and repair using materials from their environment or from other robots."

The robots are made from "truss links" — six-sided elongated rods with magnetic connectors that can contract and expand with other modules.

These modules can be assembled and disassembled as well. The magnets enable the robots to form increasingly complex structures in what their makers hope can be a "self-sustaining machine ecology."

Related: Watch humanlike robot with bionic muscles dangle as it twitches, shrugs and clenches its fists in creepy video

There are two rules for robot metabolism, the scientists said in the study. First, a robot must grow completely on its own, or be assisted by other robots with similar components. Second, the only external provisions granted to the truss links are materials and energy. Truss links use a mix of automated and controlled behaviors. Shape-shifting, cannibalizing robots

'Bad sci-fi scenarios'

In a controlled environment, scientists laid truss links across an environment to observe how the robot connects with other modules.

The researchers noted how the truss links first assembled themselves in 2D shapes but later integrated new parts to become a 3D tetrahedron that could navigate the uneven testing ground. The robot did this by integrating an additional link to use as a walking stick, the researchers said in the study.

"Robot minds have moved forward by leaps and bounds in the past decade through machine learning, but robot bodies are still monolithic, unadaptive, and unrecyclable. Biological bodies, in contrast, are all about adaptation — lifeforms can grow, heal and adapt," study co-lead author Hod Lipson, chair of the department of mechanical engineering at Columbia University, said in the statement.

"In large part, this ability stems from the modular nature of biology that can use and reuse modules (amino acids) from other lifeforms," Lispon added. "Ultimately, we'll have to get robots to do the same — to learn to use and reuse parts from other robots."

The researchers said they envisioned a future in which machines can maintain themselves, without the assistance of humans. By being able to grow and adapt to different tasks and environments, these robots could play important roles in.disaster recovery and space exploration, for example.

"The image of self-reproducing robots conjures some bad sci-fi scenarios," Lipson said. "But the reality is that as we hand off more and more of our lives to robots, from driverless cars to automated manufacturing, and even defense and space exploration. Who is going to take care of these robots? We can't rely on humans to maintain these machines. Robots must ultimately learn to take care of themselves."

Students have built a hybrid drone that can seamlessly transition from flying in the air to swimming in water.

The students developed a working prototype of the hybrid drone for a bachelor's thesis at Aalborg University in Denmark, and recently shared a video of the drone in action.

In the video, the drone takes off next to a large pool of water and then quickly dives underwater. It then moves around beneath the surface for a few seconds before shooting straight out of the water to fly once again. The video shows the drone repeating the trick several times from different angles.

Andrei Copaci, Pawel Kowalczyk, Krzysztof Sierocki and Mikolaj Dzwigalo, who are all studying applied industrial electronics, achieved this remarkable air-to-water transition by using variable pitch propellers, which have blades that can rotate at different angles to match the two different environments.

"The development of an aerial underwater drone marks a major step forward in robotics, showing that a single vehicle can operate effectively in both air and water thanks to the use of variable pitch propellers," the students told Live Science in a joint email.

This isn't the first air-water hybrid drone to be built. Researchers at Rutgers University in New Jersey developed a hybrid prototype that could perform a similar action in 2015, while Chinese scientists showed off a drone transitioning from air to water in 2023.

Related: Best drones for wildlife and landscapes in 2025 — Explore and capture the world from above

The students designed, built and tested their drone over two semesters at their university, according to a LinkedIn post by Petar Durdevic, an associate professor who leads the Offshore Drones and Robots research group at Aalborg University.

They began by creating a model of the drone and designing the variable pitch propeller system. The angle of the blades, or propeller pitch, is higher when flying to create more air flow, while lower in the water to minimize drag and increase efficiency. The propellers are also capable of providing negative thrust to increase maneuverability underwater, the students said.

The team used a 3D printer and a computer numerical control machine — another piece of automated manufacturing equipment — to get the parts they needed for the build, and programmed the drone with custom software. Finally, they moved on to testing.

"We were surprised how seamlessly the drone transitions from water to air," the students said.

The new drone is just a single prototype, but this kind of technology has a variety of potential real-world applications, from emergency response to warfare. "A few of the applications are military, vessel inspections, marine exploration, search and rescue," the students said.

There are many weird and wonderful humanoid robots out there, but one of the most eye-catching machines launched this year can change its own battery pack — making it capable of running autonomously for 24 hours a day, seven days a week.

The Walker S2 robot, made by the Chinese company UBTECH, is 5 foot 3 inches (162 centimeters) tall and weighs 95 pounds (43 kilograms) — making it the size and weight of a small adult.

Using a 48-volt lithium battery in a dual-battery system, the robot can walk for two hours or stand for four hours before its power runs out. The battery takes 90 minutes to fully recharge once depleted.

Its most interesting feature — which UBTECH representatives say is a world first — is that instead of relying on a human operator to remove and recharge its battery pack, the machine can perform this task entirely on its own.

In new promotional footage published July 17 on YouTube, the Walker S2 robot is seen approaching a battery charging station to swap out its battery supply. Facing away from the station, it uses its arms to remove the battery pack fitted into its back and places this into an empty slot to recharge. It then removes a fresh battery pack from the unit and inserts it into its port.

The robot will swap out its own battery in the event that one of its batteries runs out of power. It is also capable of detecting how much power it has left and decides whether it is best to swap out one of its batteries or charge based on the priority of its tasks, company representatives said, as reported by the Chinese publication CnEVPost.

The Walker S2, which is designed to be used in settings like factories or as a human-like robot to meet and greet customers at public venues, has 20 degrees of freedom (the number of ways that joints or mechanisms can move) and is also compatible with Wi-Fi and Bluetooth.

A first-of-its-kind robot elephant can pick flowers and go bowling thanks to 3D-printed technology that mimics biological tissues, according to a new study.

Researchers in Switzerland created a programmable lattice structure that can take on a variety of different forms, giving the robot a flexible synthetic trunk that can handle delicate tasks and rigid bone-like support in its legs for more natural movement.

The researchers demonstrated the benefits of their geometric lattice design by showing the robot elephant delicately grasping a flower with its trunk and kicking a bowling ball. In a video shared by the researchers, the robot elephant kicked a small bowling ball at 10 pins — and managed to knock down seven of them.

The lattice is made from foam and contains many individual units, or cells, that the researchers can program into different shapes and positions. The team claims that their tech can have more than one million different configurations that combine to make "infinite" geometric variations, allowing for the design of lightweight and adaptable robots, according to a new study published Wednesday (July 16) in the journal Science Advances.

"We used our programmable lattice technique to build a musculoskeletal-inspired elephant robot with a soft trunk that can twist, bend and rotate, as well as more rigid hip, knee, and foot joints," study first author Qinghua Guan, a postdoctoral researcher in the Computational Robot Design and Fabrication Lab at the EPFL university in Switzerland, said in a statement. "This shows that our method offers a scalable solution for designing unprecedentedly lightweight, adaptable robots."

Related: Intrepid baby-faced robot dons a jetpack for its next adventure — becoming the first humanoid robot to fly

Most robots, including the most advanced humanoid robots, are pretty clunky and awkward in the way that they move compared with humans and other animals. Our varied movement comes from a network of muscles, tendons, ligaments and bones, often all working together, which is a complexity that's difficult to replicate in robots, according to the statement.

An elephant's trunk is a particularly complex piece of evolutionary engineering. Each trunk contains around 90,000 bundles of muscle fibers, called muscle fascicles, that allow elephants to use them like a multipurpose tool.

The 3D printed lattice gets its complexity from its individual cell units. These cells have two main geometric shapes that provide different levels of stiffness, deformation and load-bearing properties. Furthermore, researchers can create what are essentially hybrid shapes on a spectrum between the two main shapes, allowing for even more variation, according to the statement.

"This approach enables the continuous spatial blending of stiffness profiles and allows for an infinite range of blended unit cells," study co-author Benhui Dai, a doctoral student in the lab at EPFL, said in the statement. "It's particularly suited for replicating the structure of muscular organs like an elephant trunk."

Scientists at MIT have developed a novel vision-based artificial intelligence (AI) system that can teach itself how to control virtually any robot without the use of sensors or pretraining.

The system gathers data about a given robot’s architecture using cameras, in much the same way that humans use their eyes to learn about themselves as they move.

This allows the AI controller to develop a self-learning model for operating any robot — essentially giving machines a humanlike sense of physical self-awareness.

Researchers achieved this breakthrough by creating a new control paradigm that uses cameras to map a video stream of a robot’s "visuomotor Jacobian field," a depiction of the machine’s visible 3D points, to the robot’s actuators.

The AI model can then predict precision-motor movements. This makes it possible to turn non-traditional robot architectures, such as soft robotics and those designed with flexible materials, into autonomous units with only a few hours of training.

"Think about how you learn to control your fingers: you wiggle, you observe, you adapt,” explained Sizhe Lester Li, a PhD student at MIT CSAIL and lead researcher on the project, in a press release. “That’s what our system does. It experiments with random actions and figures out which controls move which parts of the robot."

Related: Scientists burned, poked and sliced their way through new robotic skin that can 'feel everything'

Typical robotics solutions rely on precision engineering to create machines to exact specifications that can be controlled using pre-trained systems. These can require expensive sensors and AI models developed with hundreds or thousands of hours of fine-tuning in order to anticipate every possible permutation of movement. Gripping objects with handlike appendages, for example, remains a difficult challenge in the arenas of both machine engineering and AI system control.

Understanding the world around you

Using the "Jacobian field" mapping camera solution, in contrast, provides a low-cost, high-fidelity solution to the challenge of automating robot systems.

The team published its findings June 25 in the journal Nature. In it, they said the work was designed to imitate the human brain’s method for learning to control machines.

Our ability to learn and reconstruct 3D configurations and predict motion as a function of control is derived from vision alone. According to the paper, “people can learn to pick and place objects within minutes” when controlling robots with a video game controller, and “the only sensors we require are our eyes.”

The system’s framework was developed using two to three hours of multi-view videos of a robot executing randomly generated commands captured by 12 consumer-grade RGB-D video cameras.

This framework is made up of two key components. The first is a deep-learning model that essentially allows the robot to determine where it and its appendages are in 3-dimensional space. This allows it to predict how its position will change as specific movement commands are executed. The second is a machine-learning program that translates generic movement commands into code a robot can understand and execute.

The team tested the new training and control paradigm by benchmarking its effectiveness against traditional camera-based control methods. The Jacobian field solution surpassed those existing 2D control systems in accuracy — especially when the team introduced visual occlusion that caused the older methods to enter a fail state. Machines using the team’s method, however, successfully created navigable 3D maps even when scenes were partially occluded with random clutter.

Once the scientists developed the framework, it was then applied to various robots with widely varying architectures. The end result was a control program that requires no further human intervention to train and operate robots using only a single video camera.

U.S. tech startup Skydweller Aero has teamed up with Thales, a French electronics company specializing in defense systems, to develop a new maritime surveillance drone that can stay aloft far longer than existing machines.

Skydweller powers itself purely from solar energy and aims to be capable of continuous flight. The initial flight milestone will be for it to remain aloft for 90 days, but ultimately it has the potential to fly for much longer.

The solar energy that powers the Skydweller is captured by over 17,000 individual solar cells, spread across approximately 2,900 square feet (270 square meters) of wing surface — across a wingspan of 236 feet (72 m), 25 feet (7.6 m) longer than a Boeing 747. In ideal conditions, the solar cells can generate up to 100 kilowatts of power for the aircraft.

During daylight hours, solar energy is used to maintain flight, power the onboard avionics and charge batteries. The Skydweller has over 1,400 pounds (635 kilograms) of batteries, which are used to power the aircraft through the night. This will allow Skydweller to maintain almost continuous flight.

The Skydweller typically flies at an altitude between 24,600 and 34,400 feet (7,500 and 10,500 meters), but can fly as high as 44,600 feet (13,600m) during the day, before dropping by 4,900 to 9,800 feet (1,500 to 3,000m)at night, as this minimizes power consumption.

Despite its similar wingspan to a long-range commercial airliner, Skydweller weighs 160 times less than a "jumbo jet" — 2.5 metric tons at maximum capacity versus 400 tons for the 747 at full payload.

Related: China has developed the largest drone carrier in the world — and it's getting ready for takeoff

Solar-powered aircraft are not completely new, but some designs have suffered structural problems, including catastrophic failure mid-flight when climbing or descending through medium altitudes (approximately 6,500-32,800 feet, or 2,000-10,000 m).

The Skydweller has been specifically designed to operate in this altitude range, using automatic gust-load alleviation software in the flight control system to reduce the aerodynamic loads caused by turbulence. It has also been constructed from carbon fiber and can carry up to 800 pounds (362 kg) of payload.

Continuous surveillance by sky

Operating an aircraft continuously and reliably for up to 90 days necessitates a quadruple-redundant flight control system and vehicle management system (VMS). Should one of the onboard systems fail, a backup system can take over to maintain the flight.

Self-healing algorithms within the VMS allow any failed strings (coding in an algorithm) to be autonomously shut down, corrected and resurrected during flight, thereby allowing the aircraft to return to quadruple redundancy, according to information published by company representatives. This enables the aircraft to consistently maintain flight.

Although the onboard batteries, once sufficiently charged, can maintain flight during the night, their capacity will degrade over time, which could limit the maximum patrol duration of the aircraft. Skydweller’s reliance on solar power to maintain flight means that its patrols must also avoid areas of limited sunlight, such as polar regions during winter.

Skydweller Aero has recently partnered with Thales to equip Skydweller with a radar surveillance system designed for maritime patrol operations. Further test flights are planned, with the goal of extending the maximum flight duration. Even so, this is a massive step forward in solar-powered flight, especially for long-term surveillance monitoring.

Scientists have developed a new type of electronic "skin" that could give robots the ability to "feel" different tactile sensations like pokes, prods and temperature changes — and even the feeling of being stabbed.

The skin is made from an electrically conductive, gelatin-based material that can be molded into different shapes. When equipped with a special type of electrode, the material can detect signals from hundreds of thousands of connective pathways that correspond to different touch and pressure sensations.

The scientists said the material could be used in humanoid robots or human prosthetics where a sense of touch is vital, in addition to having broader applications in the automotive sector and in disaster relief. They published their findings June 11 in the journal Science Robotics.

Tactile sensing has emerged as the next big milestone for robotics, as scientists look to build machines that can respond to the world in a manner akin to human sensitivity.

Electronic skins typically work by converting physical information — like pressure or temperature — into electronic signals. In most cases, different types of sensors are needed for different types of sensation; for example, one to detect pressure, another to detect temperature and so on.

However, signals from these different sensors can interfere with each other, and the materials they are embedded in — traditionally soft silicones or stretchy, rubber-like materials called elastomers — are easily damaged, the scientists said.

Related: Self-healing 'living skin' can make robots more humanlike — and it looks just as creepy as you'd expect

This new electronic skin uses a single type of "multi-modal" sensor that is capable of detecting different types of stimuli like touch, temperature and damage.

While it's still challenging to reliably separate and pinpoint the cause of each signal, multi-modal sensing materials are easier to fabricate and more robust, the scientists said. They're also less expensive to produce, making them suitable and cost-effective for widespread use.

That's handy

To test their synthetic flesh, the researchers melted down a soft, stretchy and electrically conductive gelatin-based hydrogel, and cast it into the shape of a human hand. They then equipped the hand with different electrode configurations to see which captured the most useful data from physical interactions, subjecting it to a series of tests to find out.

This rather brutal process involved blasting it with a heat gun, poking it with their fingers and a robotic arm, and cutting it open with a scalpel.

In total, the researchers said they collected more than 1.7 million pieces of information from the skin's 860,000-plus conductive pathways. They used data gathered from these tests to train a machine learning model that, if integrated into a robotic system, could enable it to recognize different types of touch.

"We're not quite at the level where the robotic skin is as good as human skin, but we think it’s better than anything else out there at the moment," study co-author Thomas George Thuruthel, a lecturer in robotics and artificial intelligence (AI) at University College London (UCL), said in a statement.

"Our method is flexible and easier to build than traditional sensors, and we're able to calibrate it using human touch for a range of tasks."

Scientists have developed a new kind of "electronic ink" that can be used to print electronic circuits capable of switching between rigid and soft states when heated.

The technology could pave the way for next-generation electronic devices that change shape or stiffness depending on how and where they’re used, from medical implants that soften inside the body to flexible robotics.

The ink combines gallium, a metal that's solid at room temperature but melts just below body temperature — 98.6 degrees Fahrenheit (37 degrees Celsius) — with a polymer-based solvent that breaks down when gently heated. The result is a stable, printable substance that becomes conductive after heating and can change its stiffness in response to temperature.

The findings were published May 30 in the journal Science Advances.

"This opens up new possibilities for future personal electronics, medical devices and robotics," study co-author Jae-Woong Jeong, professor of electrical engineering at the Korea Advanced Institute of Science and Technology (KAIST), said in a statement.

Related: Unique transistor 'could change the world of electronics' thanks to nanosecond-scale switching speeds and refusal to wear out

Most electronics today fall into one of two camps: rigid devices like smartphones and laptops, which offer performance and durability at the cost of flexibility; or soft systems like wearables, which are more comfortable to wear but can be harder to fabricate precisely or integrate with more complex components.

From hardware to soft-wear

So-called variable-stiffness electronics aim to bridge this gap, enabling devices to shift between hard and soft states as needed.

Gallium has long attracted interest in this area because of how differently it behaves in solid and liquid forms. But using it for printed electronics has proven difficult due to its high surface tension and tendency to oxidize when exposed to air, forming a kind of crust that prevents it from bonding or spreading properly.

To address this, the researchers developed a process for dispersing microscopic gallium particles into a polymer matrix — essentially a flexible, ink-like base — using a solvent called dimethyl sulfoxide (DMSO).

When the printed circuit is gently heated, the solvent breaks down and creates a slightly acidic environment. This strips away the oxide layer from the gallium particles, enabling them to melt and merge to form conductive pathways.

The resulting ink can be used to print features as small as 50 micrometers (0.002 inches or 0.005 centimeters) — thinner than a human hair — and can alternate between plastic-like hardness and rubbery softness as needed. The researchers said that the material became more than 1,400 times softer when warmed during tests.

The team built two working devices to demonstrate how the bendy technology might be used. One was a wearable health device that behaves like a rigid portable electronic at room temperature, then softens on contact with skin to improve comfort. The other was a flexible brain implant that remained rigid during surgery so it could be precisely inserted, and then softened once inside the brain to help reduce irritation and inflammation.

The ink can be used with common manufacturing techniques such as screen printing and dip coating, meaning it could be used in larger-scale or 3D-printed electronics in the future, the researchers said.

"The core achievement of this research lies in overcoming the longstanding challenges of liquid metal printing through our innovative technology," Jeong said in the statement. "By controlling the ink's acidity, we were able to electrically and mechanically connect printed gallium particles, enabling the room-temperature fabrication of high-resolution, large-area circuits with tunable stiffness."

Researchers have demonstrated the first successful vertical liftoff of a humanoid robot — a major step toward flying robots that mimic human capabilities.

The new robot, dubbed iRonCub MK3, uses four thrusters to achieve vertical thrust, two built into its arms and two in a jetpack mounted on its back.

In footage released by the Italian Institute of Technology (IIT) to YouTube on 18 June, the mask-donning robot is held from a suspender and then activates, lifting off roughly 20 inches (50 centimeters) from the ground using its thrusters.

The iCub robot that IIT's humanoid is based on is child-like in both size and appearance. It measures 3 feet (1 meter) tall while weighing 49 pounds (22 kilograms), adopting child-like dimensions, and donning a highly recognizable smooth and shiny "baby face" head.

First launched in 2009 by the RobotCub Consortium of European universities, it is designed to be a testbed for research into human cognition and artificial intelligence (AI) and has been used in countless projects over the last 16 years.

In a corresponding study describing the robot, published June 1 on the arXiv preprint server, the IIT engineers say their augmented version of the robot is the first such jet-powered flying humanoid robot designed to take off. They added this work highlights the new possibilities that a flying, humanoid robot presents.

Related: Robots: Facts about machines that can walk, talk or do tasks that humans can't (or won't)

In contrast to many other robotic platforms currently in development, which typically focus exclusively on ground-based locomotion and manipulation, adding aerial mobility may pave the way for assistance in search and rescue missions, traversing collapsed buildings and flooded areas, or other terrain inaccessible to humans, the scientists said in the study.

Combining flight with the ability to interact with its environment, a robot of this kind could clear debris and assist in locating or even extracting victims.

The authors of the study also point out that a flying humanoid robot could carry out repairs on structures difficult for human beings to access, like the undersides of bridges. They could even be deployed in hazardous environments where people can't safely operate, like the site of a nuclear or chemical leaks.

The world's first jetpack-powered flying robot

iRonCub is based on IIT's iCub3 platform. This system is designed to allow a robot to emulate a number of human capabilities as it's being remotely controlled by an operator wearing a virtual reality (VR) helmet-like heads-up display (HUD) and other specialized equipment.

To equip the iCub for flight, the team added load-bearing mounting brackets to secure the jetpack and a new titanium spine to support its weight. Because previous versions of the iRonCub had issues with catching fire from the thrusters' exhaust, heat-resistant panels and covers were added to the iCub chassis. The thrusters' tilt angle was also adjusted to help with in-flight stability and to angle the exhaust away from the robot's body.

For this version, the robot's forearms and hands were removed to make space for the arm-mounted thrusters, but the team says the forearms are still fully compatible and will be readded in later iterations.

To ensure the chassis could handle the stress of takeoff and flight, the team performed a Finite Element Method (FEM) analysis — a problem-solving technique that centers around breaking a structure down into smaller interconnected components. As a result, they applied axial loads of 750 newtons (three times the peak thrust of each turbine) to the brackets of the jet engines.

One of the major challenges of achieving flight in a humanoid robot of this type is accounting for all of the joints and actuation points and adjusting them, as well as the thrusters, in a coordinated way, so the robot doesn't end up flying off course or upside down.

The engineers managed this with what's called a Linear Parameter Varying Model Predictive Control (MPC), a complex flight controller that's constantly computing the optimal joint position and jet throttle.

The scientists extensively tested the system first in computer simulations before then launching real-world tests, in large part due to the safety risks of operating thrusters with exhaust temperatures in excess of 1,112 degrees Fahrenheit (600 degrees Celsius).

After successfully demonstrating a launch with a predefined trajectory and accurate tracking, the team now plans to focus its efforts on improving the flight controller to handle external disturbances that may occur during a real flight scenario, like wind or other inclement weather.

Lifelike humanoid robots have competed in the world's first humanoid robot combat competition, with four Chinese teams pitting advanced fighting robots against each other.

The China Media Group (CMG) "World Robot Tournament — Mecha Combat Arena'" competition was held in Hangzhou, China, May 25 and broadcast nationwide on China's state-run CCTV-10 science channel. During the tournament, viewers watched Unitree's G1 humanoid robots spar for the first time — throwing punches and kicks and dodging incoming attacks.

Four teams — black, pink, green and red — took part in the competition, with each controlling a robot using unique tactics. For example, the green team, led by electrical engineer Hu Yunqian (nicknamed "Power Core Guardian") used a "steady pressure, inch-by-inch advance" approach to defeat opponent robots.

Throughout the fight, the team chose to pilot the robot remotely using a game pad because this method had a lower latency — the lag between when a command is issued and when the robot responds — than other control modes, such as artificial intelligence (AI) voice control and motion sensing, Unitree representatives explained.

In contrast, the pink team leader, Jiao Tianqi (nicknamed "The Silk Artisan"), ran a "muscle memory precision" strategy focused on leveraging reflexes and overwhelming opponents with sheer attack speed.

Metalweight boxing

The G1 humanoid robot is capable of moving across numerous axes and with high precision, Unitree representatives said in a statement.

In addition to throwing punches using its shoulder joints, it can swing its hips and knees to kick opponents. Unitree representatives also stated that the G1's arms have seven degrees of freedom — more than other robots and the same as a human arm. These break down into three possible rotations of the shoulder and wrist, respectively, as well as the opening and closing motion of the elbow.

Humanoid robots tend to need special algorithms and sensors to keep them from toppling over when carrying out complex movements. Whereas most humans learn to balance from a young age, robots need precise instructions to maintain their balance when performing tasks.

The G1 highlights the leaps researchers have made in getting robots to not only punch without falling over but also stay upright while kicking and dodging attacks. In combat, the robots experienced intense periods of physical strain, which they were engineered to withstand.

Unitree prepared the G1 to balance through these fights using AI reinforcement training, in which optimal behaviors are learned through exposure to massive amounts of data. The resulting algorithm was used in tandem with real-time balance sensors to keep the G1 upright with or without human control.

To keep the robot stable in combat situations, Unitree ran additional rounds of training in which the robot was hit from different angles and made to stay upright. They also equipped it with specific learned movements that helped it to balance, such as lowering its center of gravity through its legs or swinging its arms to stop itself from falling over.

Unitree stated that in future tournaments, it will deploy a robot with a more capable motion-sensing control system, for more "immersion" in combat.

Earlier this year, a robot completed a half-marathon in Beijing in just under 2 hours and 40 minutes. That's slower than the human winner, who clocked in at just over an hour — but it's still a remarkable feat. Many recreational runners would be proud of that time. The robot kept its pace for more than 13 miles (21 kilometers).

But it didn't do so on a single charge. Along the way, the robot had to stop and have its batteries swapped three times. That detail, while easy to overlook, speaks volumes about a deeper challenge in robotics: energy.

Modern robots can move with incredible agility, mimicking animal locomotion and executing complex tasks with mechanical precision. In many ways, they rival biology in coordination and efficiency. But when it comes to endurance, robots still fall short. They don't tire from exertion — they simply run out of power.

As a robotics researcher focused on energy systems, I study this challenge closely. How can researchers give robots the staying power of living creatures — and why are we still so far from that goal? Though most robotics research into the energy problem has focused on better batteries, there is another possibility: Build robots that eat.

Robots move well but run out of steam

Modern robots are remarkably good at moving. Thanks to decades of research in biomechanics, motor control and actuation, machines such as Boston Dynamics' Spot and Atlas can walk, run and climb with an agility that once seemed out of reach. In some cases, their motors are even more efficient than animal muscles.

But endurance is another matter. Spot, for example, can operate for just 90 minutes on a full charge. After that, it needs nearly an hour to recharge. These runtimes are a far cry from the eight- to 12-hour shifts expected of human workers — or the multiday endurance of sled dogs.

Related: Amazon's new warehouse robot has a 'sense of touch' that could see it replace human workers

The issue isn't how robots move — it's how they store energy. Most mobile robots today use lithium-ion batteries, the same type found in smartphones and electric cars. These batteries are reliable and widely available, but their performance improves at a slow pace: Each year new lithium-ion batteries are about 7% better than the previous generation. At that rate, it would take a full decade to merely double a robot's runtime.

Animals store energy in fat, which is extraordinarily energy dense: nearly 9 kilowatt-hours per kilogram. That's about 68 kWh total in a sled dog, similar to the energy in a fully charged Tesla Model 3. Lithium-ion batteries, by contrast, store just a fraction of that, about 0.25 kilowatt-hours per kilogram. Even with highly efficient motors, a robot like Spot would need a battery dozens of times more powerful than today's to match the endurance of a sled dog.

And recharging isn't always an option. In disaster zones, remote fields or on long-duration missions, a wall outlet or a spare battery might be nowhere in sight.

In some cases, robot designers can add more batteries. But more batteries mean more weight, which increases the energy required to move. In highly mobile robots, there's a careful balance between payload, performance and endurance. For Spot, for example, the battery already makes up 16% of its weight.

Some robots have used solar panels, and in theory these could extend runtime, especially for low-power tasks or in bright, sunny environments. But in practice, solar power delivers very little power relative to what mobile robots need to walk, run or fly at practical speeds. That's why energy harvesting like solar panels remains a niche solution today, better suited for stationary or ultra-low-power robots.

Why it matters

These aren't just technical limitations. They define what robots can do.

A rescue robot with a 45-minute battery might not last long enough to complete a search. A farm robot that pauses to recharge every hour can't harvest crops in time. Even in warehouses or hospitals, short runtimes add complexity and cost.

If robots are to play meaningful roles in society assisting the elderly, exploring hazardous environments and working alongside humans, they need the endurance to stay active for hours, not minutes.

New battery chemistries such as lithium-sulfur and metal-air offer a more promising path forward. These systems have much higher theoretical energy densities than today's lithium-ion cells. Some approach levels seen in animal fat. When paired with actuators that efficiently convert electrical energy from the battery to mechanical work, they could enable robots to match or even exceed the endurance of animals with low body fat. But even these next-generation batteries have limitations. Many are difficult to recharge, degrade over time or face engineering hurdles in real-world systems.

Fast charging can help reduce downtime. Some emerging batteries can recharge in minutes rather than hours. But there are trade-offs. Fast charging strains battery life, increases heat and often requires heavy, high-power charging infrastructure. Even with improvements, a fast-charging robot still needs to stop frequently. In environments without access to grid power, this doesn't solve the core problem of limited onboard energy. That's why researchers are exploring alternatives such as "refueling" robots with metal or chemical fuels — much like animals eat — to bypass the limits of electrical charging altogether.

An alternative: Robotic metabolism

In nature, animals don't recharge, they eat. Food is converted into energy through digestion, circulation and respiration. Fat stores that energy, blood moves it and muscles use it. Future robots could follow a similar blueprint with synthetic metabolisms.

Some researchers are building systems that let robots "digest" metal or chemical fuels and breathe oxygen. For example, synthetic, stomachlike chemical reactors could convert high-energy materials such as aluminum into electricity.

This builds on the many advances in robot autonomy, where robots can sense objects in a room and navigate to pick them up, but here they would be picking up energy sources.

Other researchers are developing fluid-based energy systems that circulate like blood. One early example, a robotic fish, tripled its energy density by using a multifunctional fluid instead of a standard lithium-ion battery. That single design shift delivered the equivalent of 16 years of battery improvements, not through new chemistry but through a more bioinspired approach. These systems could allow robots to operate for much longer stretches of time, drawing energy from materials that store far more energy than today's batteries.

In animals, the energy system does more than just provide energy. Blood helps regulate temperature, deliver hormones, fight infections and repair wounds. Synthetic metabolisms could do the same. Future robots might manage heat using circulating fluids or heal themselves using stored or digested materials. Instead of a central battery pack, energy could be stored throughout the body in limbs, joints and soft, tissuelike components.

This approach could lead to machines that aren't just longer-lasting but more adaptable, resilient and lifelike.

The bottom line

Today's robots can leap and sprint like animals, but they can't go the distance.

Their bodies are fast, their minds are improving, but their energy systems haven't caught up. If robots are going to work alongside humans in meaningful ways, we'll need to give them more than intelligence and agility. We'll need to give them endurance.

This edited article is republished from The Conversation under a Creative Commons license. Read the original article.

Amazon has announced a new robot with a sense of touch, enabling it to pick up and stow around three-quarters of the items found in its warehouses.

The new robot, called Vulcan, can accurately gauge the pressure required to grab and move warehouse items based on their size, dimensions and density, according to the e-commerce giant.

"Vulcan represents a fundamental leap forward in robotics," Aaron Parness, Amazon's director of applied science, said in a statement. "It's not just seeing the world, it's feeling it, enabling capabilities that were impossible for Amazon robots until now."

To achieve human-like touch sensitivity, Vulcan deploys a wand with both an item-identifying camera and a suction cup. Vulcan uses the wand to reach into warehouse stowing bins, move items around and identify the one it's searching for. This wand is at the end of a two-part robotic arm that can grip and pull items into or out of Amazon's inventory pods.

Related: Swarm of 30 robots can 'flow like water' and harden up to support the weight of a person

The wand's suction cup pulls items forward to the gripper, which functions as a conveyor belt to further move items in and out of bins. By feeding data into its artificial intelligence system, Vulcan learns from each item interaction and adjusts future engagements based on what it has learned. Vulcan can also sense when it can't move an item and can ask for human assistance.

Vulcan is currently in operation at Amazon's Spokane, Washington, and Hamburg, Germany, fulfillment centers.

Boston Dynamics, Fetch Robotics (which was acquired by Zebra Technologies in 2021) and RightHand Robotics have all faced difficulties in training robots to mimic human levels of touch sensitivity. This is because robots would inaccurately identify the durability of items and either crush them or not apply enough pressure to grab and transport them properly, according to Parness.

Over the past 12 years, Amazon has deployed 750,000 robots in its warehouses across the globe — a number that nearly matches the company's 1 million human employees.

Amazon said that, in addition to improving efficiency, this robotic fleet will reduce the ergonomic stress of its warehouse employees by eliminating the need to use stepladders or bend down to reach items. The company also claims the scaling of Vulcan across its warehouses will create a new category of jobs for humans, including robotic floor managers and employees to oversee and ensure the quality of the robotic fulfilment of orders.

Yet Amazon has another ambition in mind for Vulcan and its robotic fleet: reducing the company's reliance on human labor. An internal document acquired by Business Insider said Amazon's robots are "critical to flattening Amazon's hiring curve over the next ten years" to save the company as much as $10 billion annually by 2030.

Swarms of hive mind-like robots can be induced to behave more like a material, new research has shown.

The robots in question are (mostly) cylindrical, with a diameter of just 2.75 inches (70 millimeters). The researchers at UC Santa Barbara and TU Dresden used 3D printing technology to manufacture the pieces from polylactic acid.

Each robot has a magnet in its base, yellow gears made out of polylactic acid, and a center with a ring of gears. Collectively, they can perform structure-forming and healing, and support 700 newtons (500 times the weight of a robot). The scientists outlined the structure in a study published Feb. 20 in the journal Science.

The robots move thanks to magnets located in the base. These allow the robots to "naturally stick together and remain cohesive," according to the lead author of the study Matthew Devlin, Ph.D candidate at UCSB, in an email to Live Science.

"These robots are inspired by developing embryonic cells, which are cohesive and push and pull against each other to form the complex structures of life."

The gears apply a force pattern that mirrors the behaviors of embryonic tissue cells, for the overall shape and structure to change within a hexagonal lattice structure. Footage provided by the team shows their work in action, with the robots self-organizing into a pattern similar to that of the cells within a honeycomb.

Related: Self-healing 'living skin' can make robots more humanlike — and it looks just as creepy as you'd expect

Eight motorized gears enable intracellular forces to become tangential forces that move each unit. This allows the robots to push, pull, and move around one another no matter how narrow a space may be.

One of the team’s primary research goals was creating a material that could be rigid when the situation required it to be, but also adapt itself to assume a softer structure. They also set the goal of wanting the robots to not only take a shape they could maintain but also "selectively flow themselves into a new shape."

An embryonic technology

A former professor at UC Santa Barbara, Otger Campàs, previously studied how embryos are physically shaped. The work by Campas, currently the director of the Physics of Life Excellence Cluster at TU Dresden, helped inform the team of researchers at his former institution when developing the robots.

Campas’s lab at UCSB had discovered a unique property of embryos — the ability to "sculpt themselves."

"To sculpt themselves, cells in embryos can make the tissues switch between fluid and solid states; a phenomenon known as rigidity transitions in physics," Campas said in an article featuring the research, titled “Material-like robotic collectives with spatiotemporal control of strength and shape” published in the journal Science.

Though the team successfully recreated the motions and state transitions of embryonic cells in its robots, Devlin shared with the Live Science team that what he found most fascinating about the experiment was not the change of state from solid to liquid, but the change of strength.

"By fluctuating the magnitude of the forces the gears are applying to each other, the robots can change whether they behave more like a liquid or solid," Devlin said. He went on to say that his favorite demonstration of this feature occurs when the researcher can stand on the robots.

By completing of a proof-of-concept swarm, and following various simulations conducted by postdoctoral fellow Sangwoo Kim, the team was able to achieve its initial goal of creating and controlling robots to make them act like a material.

The team hopes to explore the scalability of the system, and wish to further study the phase transitions in active matter rather than through the robots. This will help define their hypotheses in order to apply the knowledge accrued in their upcoming research.

Robots are machines that can automatically perform both simple and complex actions. They are often programmed by a computer, but there's no single definition of a robot. Robots include a wide variety of machines, such as the giant arms used to make cars, automatic vacuum cleaners, humanoid machines that look like people and do backflips, and robotic dogs.

As technology develops and gets more powerful, we're seeing more advanced robots. In fact, some robots powered by artificial intelligence (AI) can have normal conversations.

Robots have been around for decades, and you're likely reading this article on a computer or a phone that may have been assembled by a robot.

5 fun facts about robots

• The definition of robotics is older than actual robots. In fact, the ancient Greek writer Homer described "intelligent machines" such as wheeled tripod machines in "The Iliad" around 2,800 years ago.
• Scientists have made robot "animals," that can swim like fish, hop like kangaroos, scuttle like cockroaches or scale walls.
• Robots have been designed to do almost every task people do. For instance, the SwagBot herds cows, Proteus delivers Amazon packages, and the humanoid RoboThespian mops floors and mows the lawn.
• A robot has become a citizen. The humanoid robot Sophia was granted citizenship in Saudi Arabia in October 2017.
• NASA's Voyagers 1 and 2 robotic spacecraft have left the solar system and entered interstellar space in 2012 and 2018, respectively. That means Voyager 1, at 15.6 billion miles (25.6 billion kilometers) away from Earth and counting, is the most distant man-made object in the known universe.

Everything you need to know about robots

Robot pictures

Discover more about robots

—8 of the weirdest robots in the world right now

—12 pivotal moments in the history of robotics, from Isaac Asimov to self-driving cars

—The 7 most disturbing humanoid robots

A robotics company has showcased the jerky, uncanny-valley movements of its muscular humanoid robot in a horrifying new video.

Engineers at Clone Robotics, a startup founded in Poland in 2021, are building androids that look more humanlike than any other humanoid robot built to date and mimic human movement.

The new video, released by the company on April 9, shows their translucent-white skinned "Protoclone" robot hanging from the ceiling with its legs in a plié position, while its arms, head, and hands move eerily. The robot can be seen jerking around like a marionette, shrugging its shoulders, flexing its hands into fists, moving its arms up and down, and nodding its head.

"Meet Clone's first musculoskeletal android: Protoclone, the most anatomically accurate robot in the world," company representatives wrote in the caption of the video. "Based on a natural human skeleton, Protoclone is actuated with over 1,000 Myofibers, Clone's proprietary artificial muscle technology."

In humans and animals, muscles are attached to the skeleton via tendons, which are strong connective tissues. When a muscle contracts, the contraction pulls on the tendon, which pulls on the bones, moving them around a joint.

The Protoclone robot has a realistic human-like skeleton, Clone Robotics representatives say, and is equipped with the company's artificial muscles called Myofibers, which are attached to the appropriate bones using artificial ligaments and connective tissues.

A humanlike android with humanlike movement

According to Clone Robotics' website, the robot contains all 206 human bones made from "cheap and durable polymers." The shoulder joints of the Clone, which connect the shoulder blade, collarbone, and upper arm bones, have a total of 20 degrees of freedom, which is the number of independent movements a joint can perform.

Hinge joints like the knee and elbow only have one degree of freedom each, while ball and socket joints like the hip have three degrees of freedom. Alongside the 26 degrees of freedom allowed by the hand, elbow, and wrist joints, Clone Robotics representatives claim that the Clone's upper torso alone has 164 degrees of freedom.

Related: 8 of the weirdest robots in the world right now

The Myofibers of the robot were first invented by the company in 2021, and are the "only artificial muscle in the world capable of achieving such a combination of weight, power density, speed, force-to-weight, and energy efficiency," according to the Clone Robotics website. They also state that Myofibers allow "[contraction] faster than human, skeletal muscle fibers."

The Protoclone also uses water-powered hydraulics to move its muscles, driven by a battery-powered electric pump.

"The Clone’s vascular system is the most sophisticated hydraulic powering system ever designed, with a 500 watt electric pump as compact as the human heart able to pump liquid," the Clone Robotics site reads.

The android will also be equipped with hundreds of sensors, but will not be able to feel touch or pain — only where its limbs are in reference to the rest of its body. In total, there are four depth cameras in the skull for vision, 70 inertial sensors that provide joint-level proprioception (angles and velocities) and 320 pressure sensors for muscle-level force feedback.

The full-limbed Protoclone was first revealed in February of this year in a video that went viral for its creepy movements, and the company has also previously showcased the Clone Torso in 2024, and the Clone Hand in 2022, which could rotate its thumb and even catch a ball.

The Protoclone is a prototype of the company's Clone Alpha android, which they claim will "walk naturally," perform household chores like vacuuming, laundry and meal preparation, and even "shake hands with your friends" and spout "witty dialogue."

A new type of robot inspired by an animal has been unveiled — and this time, it's a horse.

Corleo, a hydrogen-powered robot with four legs designed to be ridden by humans, has been announced by Japanese manufacturer Kawasaki Heavy Industries at a showcase event ahead of the Osaka-Kansai Expo 2025, which starts April 13. However, the robot is still in its concept phase of development; a functioning prototype has not been made public.

This robot, unlike previous animal-shaped bots such as Boston Dynamics' dog Spot, is designed for people to ride it as the machinery uses artificial intelligence (AI) and specially designed legs to navigate complex terrains.

A promotional video released by Kawasaki on April 3 shows Corleo clambering down rocky cliffs, weaving through an overgrown woodland, bounding across a snowy landscape and wading through a stream. This video is not real footage, however, as the robot has not yet been fully developed to the extent that you can ride it.

"CORLEO is a completely new means of transportation that combines Kawasaki’s motorcycle and robotics technologies," Kawasaki executive officer Takashi Torii said, as reported by Tokyo Weekender.

The robot has four legs, each of which can move independently. These legs are also equipped with inward-facing knees and rubber two-toed cloven hooves, similar to those of a goat or a deer. Company representatives hope that it can one day carry up to two people on its back through a wide range of environments.

Related: Watch this humanlike robot 'rise from the dead' with creepy speed and stability

"The rear leg unit can swing up and down independently from the front leg unit, allowing it to absorb shocks during walking and running," Kawasaki representatives explained on the Corleo concept page of their website.

"The four legs are equipped with hooves featuring a left-right divided structure made of rubber, a material that absorbs surface irregularities and is slip-resistant," according to its website. "These hooves can adapt to various terrains, including grasslands, rocky areas, and rubble fields."

Horse power

Kawasaki proposes that Corleo will have an AI system that enhances its balance and navigation, with the robot responding to the body movements of the rider — just like a real horse. The robot will also have a heads-up display in front of the rider much like a motorcycle does, showing important information.

"Equipped with an instrument panel that displays hydrogen level, route to the summit, center of gravity position, and other information. At night, it supports optimal riding by projecting markers onto the road surface to indicate the path ahead," Kawasaki representatives added.

Corleo will be driven by a 150cc engine powered by a hydrogen fuel cell. Hydrogen fuel cells work by converting chemical energy from hydrogen into electricity. Hydrogen gas is split into protons (H+) and electrons, and the electrons create electricity that can power the motor. This process produces zero emissions, with only water as a byproduct. Hydrogen fuel cells can easily be recharged by refuelling them with hydrogen gas.

This robot is still very much a concept for now, however. Kawasaki representatives suggested it may be launched by 2050, but they have not announced any solid development timeline, and there are no current plans for commercialization. The company also has not revealed any detailed technical specifications, including Corleo's speed, range of travel or battery life.

Rather than a functional prototype, a mock-up of Corleo will be featured at the Osaka-Kansai Expo, which runs from April 13 until October 13.

Scientists just showcased a humanoid robot performing a complicated side flip.

The company that makes the robot, Unitree, posted a video to Youtube showcasing its acrobatics. In the video, the silver-grey G1 crouches slightly, then launches up before rotating sideways through the air.

The robot catches itself primarily with its left leg and stabilizes almost immediately as the other foot makes contact with the floor. As impressive as it is at full speed, it's even more mesmerizing in slow motion, particularly the ease with which the robot seems to balance and right itself after landing.

Last year, the same model mastered a backflip. To teach G1 its new trick, the company mainly upgraded the artificial intelligence (AI) algorithm it used to make the software faster and more responsive, Unitree representatives told Live Science.

"The side flip was performed under reinforcement learning training," they added.

Related: Watch bipedal robots running in a more human-like way than ever thanks to major vision upgrade

Reinforcement learning is a mainstay technique used to teach robots how to navigate and interact with the physical world. This is the same technology that robotics company Figure has used to train its Figure 02 robots to move in a more humanlike way.

A robotic acrobat

The G1 can do more than acrobatics; during a martial arts demonstration, it disarmed a baton-wielding opponent. After a series of feints with its hands, the bot executed a spinning kick that sent the baton flying from its opponent's hands.

Like its human counterparts, the robot's lightweight, compact form helps it perform acrobatic moves with ease.

The G1 stands at just over 4.3 feet (1.3 meters) in height and weighs only 77 pounds (35 pounds) G1 also sports a 3D light detecting and ranging (LIDAR) and depth camera, which gives it a 360-degree view of its environment. Perhaps most importantly, it incorporates 23 degrees of freedom, a measure of the number of joints or axes of movement available to the bot.

The G1 can also walk and run at up to 2 meters per second (4.5 miles per hour or 7.2 kilometers per hour).

The company envisions robots like G1 or its successors doing everything from helping out with chores at home to performing industrial operations to assisting with hazardous rescue missions, the Unitree representative said.

A U.S. robotics company has used artificial intelligence (AI) to give its humanoid robots a more natural-looking stride, and they say it's just the beginning.

In the promotional video, the robot, called Figure 02 and manufactured by the company Figure, marches with a "human-like" gait. This is an ability it claims will help its robot to navigate the physical world more easily.

"These initial results are exciting, but we believe they only hint at the full potential of our technology," company representatives wrote in a blog post accompanying its announcement. "We're committed to extending our learned policy to handle every human-like scenario the robot might face in the real world."

Robots have been running, cartwheeling, doing backflips, breakdancing, and beating us at chess for years now. But performing tasks that appear simple to humans, such as walking in a straight line, gripping objects, tying shoelaces, and navigating social situations, have proven tough for robots to master.

Related: Chinese humanoid robot is the 'fastest in the world' thanks to its trusty pair of sneakers

The problem, known as Moravec's Paradox, emerges because computers excel at problems that require complex calculations and large datasets, but lack our real-world experience honed by millions of years of evolution. This makes robots' shuffling gaits, well, robotic at best. At worst, it gives them the appearance that they may have soiled themselves.

To tackle the robot's unnatural gait, Figure's engineers used a learning technique called reinforcement learning — placing thousands of virtual robots inside a physics simulator that recreates various terrains, thereby improving their walking through trial and error.

By rewarding the virtual robot army for natural motions, they refined their gaits to appear more human-like. With this task accomplished, they uploaded the refined "Learned Natural Walking" model to a real-world Figure 02 robot. The result is an android that can move somewhat naturally, with heel strikes, toe-offs and synchronized arm swings.

Figure's reinforcement learning technique is key to the California company's plans to roll out its robots on factory floors. It has already tested its humanoid robots in a BMW factory in 2024 and plans to introduce more this year. Meanwhile, Apptronik, a Texas-based competitor, is also commercializing its humanoid robot, Apollo, for use in Mercedes-Benz factories by the end of 2025. Agility Robotics' Digit will also be introduced into warehouses this year.

Researchers in China and Hong Kong have developed a new artificial intelligence (AI) learning framework that teaches humanoid robots to stand up from an idle position incredibly quickly regardless of position or terrain.

While the research has yet to be submitted for peer review, the team released their findings Feb. 12 on GitHub, including a paper uploaded to the arXiv preprint database, alongside a video demonstrating their framework in action.

The video shows a bipedal humanoid rising to stand after lying on its back, sitting against a wall, lying on a sofa and reclining in a chair. The researchers also tested the humanoid robot's ability to right itself on varying terrains and inclines — including a stone road, a glass slope and while leaning against a tree.

They even attempted to disrupt the robot by hitting or kicking it while it was trying to get up. In every scenario, the robot can be seen adjusting to its environment and is shown successfully standing up.

Related: 17 weird, wonderful and terrifying robots we saw at CES 2025 — from a humanlike android companion to a robotic mixologist

This remarkable ability to get knocked down and then get up again is thanks to the system called "Humanoid Standing-up Control" (HoST). The scientists achieved this with reinforcement learning, a type of machine learning where the agent (in this case the HoST framework) attempts to perform a task by trial and error. In essence, the robot takes an action, and if that action results in a positive outcome, it is sent a reward signal that encourages it to take that action again the next time it finds itself in a similar state.

Rising to the occasion

The team's system was a little more complicated than that, using four separate reward groups for more targeted feedback, along with a series of motion constraints including motion smoothing and speed limits to prevent erratic or violent movements. A vertical pull force was also applied during initial training to help direct the early stages of the learning process.

The HoST framework was originally trained in simulations using the Isaac Gym simulator, a physics simulation environment developed by Nvidia. Once the framework had been sufficiently trained on simulations, it was deployed into a Unitree G1 Humanoid Robot for experimental testing, the results of which are demonstrated in the video.

"Experimental results with the Unitree G1 humanoid robot demonstrate smooth, stable, and robust standing-up motions in a variety of real-world scenarios," the scientists wrote in the study. "Looking forward, this work paves the way for integrating standing-up control into existing humanoid systems, with the potential of expanding their real-world applicability."

Getting up might seem second nature to we humans, but it's something that humanoid robots have struggled to replicate in the past, as you can glean from a montage of robots falling over and being unable to return to an upright position. Teaching a robot to walk or run like a human being is one thing, but to be useful in the real world, they need to be able to handle challenging situations like stumbling, tripping and falling over.

A Norwegian robotics firm has unveiled a knitted-nylon-covered humanoid robot designed to complete household chores.

"Neo Gamma," built by robotics company 1X, is a bipedal android equipped to perform everyday tasks such as vacuuming, tidying clothes and making coffee.

In a promotional video released Feb 21. On YouTube, the machine is shown serving tea, fixing a wonky picture frame, carrying laundry, hoovering, wiping windows and collecting groceries, before taking a seat as its human owners eat.

Although the company has said the humanoid robot is not ready to go on sale to the public, they claim the new model has been made available for limited testing in some homes. This will enable engineers to test the robot’s navigational, speech and body language artificial intelligence (AI) features. These capabilities are being developed in-house, although ChatGPT developer OpenAI was an early investor.

"There is a not-so-distant future where we all have our own robot helper at home, like Rosey the Robot or Baymax," Bernt Børnich, the CEO of 1X, said in a statement. "But for humanoid robots to truly integrate into everyday life, they must be developed alongside humans, not in isolation."

"The home provides real-world context and the diversity of data needed for humanoids to grow in intelligence and autonomy. It also teaches them the nuances of human life — how to open the door for the elderly, move carefully around pets, or adapt to the unpredictability of the surrounding world," Børnich said.

Related: Chinese humanoid robot is the 'fastest in the world' thanks to its trusty pair of sneakers

According to 1X representatives, Neo Gamma was trained on human motion capture data, enabling it to walk with a "natural human gait and arm swings," alongside being able to squat and sit in chairs.

Its multi-joined hands use elastic motors that mimic human tendons, and it has four microphones and a speaker system integrated into its body to communicate with humans. Its knitted exterior was designed to reduce the force of potential impacts with the exterior environment and increase its overall safety.

Neo Gamma is far from the only humanoid robot that's captured the attention of businesses and the public recently. In February, Polish startup Clone Robotics announced its Protoclone robot, which twitched to life with eerily human movements.

Last year, the Californian company Figure tested its humanoid robots in a BMW factory and plans to introduce more this year. Apptronik, a Texas-based competitor, is also commercializing its humanoid robot, Apollo, for use in Mercedes-Benz factories by the end of the year. Agility Robotics' Digit, meanwhile, will also be introduced into warehouses this year.

A robotics company has built an eerily lifelike humanoid robot which they claim is the "world's first bipedal musculoskeletal android" — and the internet hates it.

Created by Polish startup Clone Robotics, the "Protoclone" robot is a faceless android with an anatomically accurate bone structure and musculature. The android's designers claim it can walk, talk and be tasked with completing menial household chores.

In a video posted on X, which quickly went viral (with over 34 million views as of Jan 21.) the robot is shown twitching to life as it's suspended from the ceiling, its limbs jerking and spasming as artificial muscles attached to specific bone points come online, enabling the skeleton to move.

Related: Chinese humanoid robot is the 'fastest in the world' thanks to its trusty pair of sneakers

"The Protoclone is a faceless, anatomically accurate synthetic human with over 200 degrees of freedom, over 1,000 Myofibers [artificial muscle fibers], and 500 sensors," Clone Robotics wrote beneath the video.

Yet while the company appears ebullient about its technological achievements, social media reactions have ranged from disdain to outright hostility.

"This is my sleep paralysis demon," one commenter wrote. "WestWorld was nonfiction," someone else added. "Can you just draw a smiley face on the head, please?'" requested another.

Rise of the robots

According to Clone Robotics' webpage, the android operates with human-like muscular, skeletal, vascular and nervous systems. It has 206 bones, the same number as adult humans, made from polymers connected by artificial muscle fibers, which it flexes using a pneumatic system (although the company says the final product will run using a hydraulics). To navigate its environment, the robot uses a system of sensors linked to four cameras.

The robot is the latest in Clone Robotics' range of biomimetic products, which also include a robotic hand and a humanoid torso. The company says that the android will be available for preorder later this year. No price is currently listed.

Protoclone is far from the only humanoid robot capturing the attention of businesses and the public. The California startup Figure tested its humanoid robots in a BMW factory in 2024, and has plans for introducing more this year. Apptronik, a Texas-based competitor, is also commercializing its humanoid robot, Apollo, for use in Mercedes-Benz factories by the end of the year. Meanwhile, Agility Robotics' Digit robot is also planned to be introduced into warehouses this year.

Dramatic footage has captured an intense battle between a robotic dog and an aerial drone as the two shoot fireworks at each other. The clash could be a glimpse into the future of warfare, although we are decades away from scenes like those in the video, an expert said.

The footage shows the drone and the dog facing off in a barren field. The drone hovers mid-air, its four propellers occasionally maneuvering the aircraft away from the dog's line of fire. Meanwhile, the dog discharges one firework after another in the direction of the drone, moving from side to side to adjust the angle of the shot.

Organizers mounted fireworks onto the back of the dog, which maneuvered itself to fire at the drone while simultaneously dodging pyrotechnics coming from the drone's landing gear. It's unclear from the footage whether either the dog or the drone were autonomous or remotely controlled, although it does appear that one of the people in the video may be flying the drone.

The video was posted on Chinese social media and sparked discussions about the future of warfare, according to the Hong Kong-based English-language newspaper South China Morning Post (SCMP).

Related: 17 weird, wonderful and terrifying robots we saw at CES 2025 — from a humanlike android companion to a robotic mixologist

It is possible that we might have robot wars in the future, and some experts argue that we already do, Missy Cummings, a professor of engineering at George Mason University in Virginia and director of Mason's Autonomy and Robotics Center, told Live Science in an email.

Advancements in artificial intelligence (AI) and other technologies have made it possible to build robots and drones that can hunt down and shoot targets with no human input. Governments around the world, including the U.S. administration, are increasingly incorporating AI into weapons and battlefield decision-making.

"Surface-to-air missiles are highly automated ground-based systems and they already shoot at drones," Cummings said. "But we are decades from anything like the video and even if that were a reality, all you would have to do is drop baby powder on all the robots and they would not be able to 'see,'" she said.

According to SCMP, the drone in the video is likely a model manufactured by DJI Agricultural, a Chinese company that designs drones for agricultural uses, such as spraying pesticides. The robotic dog is made by Chinese company Unitree Robotics, according to the newspaper.

The location where the video was taken is unknown, according to SCMP, but several people appear in the clip while the drone and dog are locked in battle.

It's unclear from the end of the footage which machine prevailed. Both finish side by side on the ground, the robot dog having seemingly run out of ammunition and the drone still firing shots away from the dog.