An AI algorithm designs and coordinates modular legs to explore billions of module connections, enabling the emergence of agile legged robot “species” and novel locomotion strategies.

The work, published in PNAS in 2026, is led by Yu, Matthews, Wang, Gu, Blackiston, Rubenstein, and Kriegman.

Experiments show metamachines achieving locomotion modes like bounding and undulating gaits, with capabilities to self-right, leap over obstacles, and perform midair maneuvers.

Unlike traditional robots, this system adapts in place when a limb is disabled or terrain is uneven, demonstrating self-adjustment for robust navigation.

Researchers first tested designs in software, then built the top three-, four-, and five-legged configurations, which could traverse gravel, grass, roots, leaves, sand, mud, and uneven bricks.

Lead author Sam Kriegman notes the approach reveals new locomotion forms and offers insights into evolution that could guide future robotic design beyond conventional configurations.

The process compresses billions of years of natural evolution into seconds, yielding robust designs that maintain movement even after limb loss or damage.

The study, titled Agile legged locomotion in reconfigurable modular robots, appears in PNAS.

Beyond robotics, the work seeks to illuminate how animals evolved to navigate diverse environments and why some species have different numbers of legs and body plans.

Northwestern researchers developed an AI-assisted, Lego-like, reconfigurable robot designed to adapt to unexpected conditions, with implications for understanding locomotion and evolution.

These AI-powered modular robots, dubbed legged metamachines, autonomously navigate varied terrains and continue moving even when damaged.

An AI-driven evolutionary algorithm simulates virtual iterations, selecting configurations that perform best on diverse terrains.