AI compressed billions of years of evolution into seconds to create 'Lego-like robots' that can recover even when they lose limbs

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.

Evolved robots are born to run, refuse to die - YouTube

Watch On

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."