Northwestern Engineers Unveil Agile, Resilient 'Legged Metamachines'

Northwestern University engineers have developed new modular legged robots, termed "legged metamachines." These innovative robots are constructed from autonomous, Lego-like modules, with each module possessing its own motor, battery, and computer. Individually, a module can perform movements such as rolling, turning, and jumping. When combined, these modules form larger machines capable of advanced agility and remarkable resilience to damage.

AI-Driven Evolution and Core Capabilities

The groundbreaking research was published on March 6, 2026, in the Proceedings of the National Academy of Sciences. A significant aspect of this development is the utilization of artificial intelligence (AI). AI was utilized to evolve novel body configurations for the robots, resulting in unique designs distinct from conventional robotic forms.

The metamachines demonstrate a wide array of capabilities. They are capable of varied movements, including undulating, bounding, and springing. Furthermore, they can self-right when overturned, hop over obstacles, and perform acrobatic maneuvers. Crucially, the modular design enables recovery from damage; individual modules can continue to function or reintegrate into the larger system if components break.

"Artificial intelligence was utilized to evolve novel body configurations for the robots, resulting in unique designs distinct from conventional robotic forms."

Pioneering Outdoor Agility and Design

The project was led by Sam Kriegman, an assistant professor at Northwestern's McCormick School of Engineering. This development marks a significant milestone: it is the first instance of an evolved robot operating outdoors and the first agile modular robot.

Engineers employed an evolutionary algorithm, simulating natural selection, to design these advanced robots. The algorithm began with half-meter-long modular legs, each containing a circuit board, battery, and motor. The primary objective for the algorithm was to optimize for efficient and versatile movement.

Real-World Performance and Damage Recovery

The optimized three-, four-, and five-legged designs were rigorously assembled and tested in diverse outdoor environments. These challenging settings included gravel, grass, tree roots, leaves, sand, mud, and uneven bricks. The metamachines remarkably demonstrated abilities such as traversing rough terrain, jumping, spinning, and self-righting without extensive pre-configuration.

A key highlight of the research is the robots' exhibited resilience to damage. The metamachines adapted and continued movement even when a leg detached. Notably, the detached module could also operate independently, underscoring the system's inherent redundancy and adaptive capabilities.

"The metamachines demonstrated abilities such as traversing rough terrain, jumping, spinning, and self-righting, even adapting and continuing movement when a leg detached."

This research builds upon previous work from Kriegman's lab involving an AI algorithm for robot design. The study received crucial funding from Schmidt Sciences AI2050 and the National Science Foundation, supporting the advancement of this cutting-edge robotic technology.