Home Engineering AI-evolved robots can survive damage and rebuild themselves in the wild

AI-evolved robots can survive damage and rebuild themselves in the wild

Called "legged metamachines," the creations are made from autonomous, Lego-like modules that snap together into an endless number of configurations. Credit: Sam Kriegman/Northwestern University.

Engineers at Northwestern University have created a new kind of robot that can adapt, recover from damage, and keep moving even after being broken apart.

The machines, called “legged metamachines,” are made from small robotic modules that snap together like Lego blocks to form many different body shapes.

Each module is a complete robot on its own. It contains its own motor, battery, computer, and simple movement system.

By itself, a module can roll, turn, and even jump. But when several modules connect together, they become far more capable.

The combined robots can crawl, hop, spin, and travel across rough outdoor terrain.

The research, published in the Proceedings of the National Academy of Sciences, shows how combining modular robotics with artificial intelligence can produce machines that are more resilient and adaptable than traditional robots.

Unlike most robots today, these metamachines are not designed entirely by human engineers. Instead, the researchers used artificial intelligence to evolve new robot body designs.

The AI used an evolutionary algorithm that mimics natural selection, similar to the way animals evolve over time.

The process begins with simple building blocks. In this case, the researchers used modular robotic legs that look like two sticks connected by a central sphere. Inside that sphere is everything the robot needs to function, including electronics, power, and a motor.

The AI then experimented with many different ways of combining the modules. It simulated thousands of possible robot designs and tested how well they moved. Designs that performed well were kept, while weaker ones were discarded. The algorithm also mixed and modified the best designs to create new generations of robots.

Over time, this process produced unusual robot shapes that human engineers might never have imagined. Some of the designs move in ways similar to animals. Certain configurations slide along the ground like seals, while others leap forward like kangaroos or bound across terrain like lizards.

Once the best designs were identified, the researchers built real versions of the robots and tested them outdoors. The metamachines successfully traveled across challenging environments, including gravel, grass, tree roots, sand, mud, and uneven bricks.

The robots could also perform impressive physical actions. They were able to jump over obstacles, spin in midair, and flip themselves upright if they were turned upside down.

One of the most remarkable features of the metamachines is their ability to survive damage. Traditional robots often stop working if a single important component breaks. But because metamachines are made of multiple robotic modules, they can continue functioning even when parts are lost.

If a leg breaks off, the rest of the robot can adjust and keep moving. The detached module does not simply become useless. Because it is still a working robot, it can roll or crawl independently and potentially reconnect with the group.

This ability to adapt makes the robots especially promising for environments where damage is likely, such as disaster zones, exploration missions, or harsh outdoor settings.

The new robots build on earlier work from the same research group, which previously developed an AI system capable of designing simple robots automatically. Those earlier robots could only perform basic movements, but they demonstrated that AI could invent working machines from scratch.

With legged metamachines, the researchers have taken a major step forward. By combining artificial intelligence with modular robotics, they have created robots that are not only strong and flexible but also capable of adapting to unexpected situations.

The scientists believe this technology could eventually lead to robots that behave less like fragile machines and more like resilient living systems capable of evolving to meet new challenges.