Most robots are fragile. They're designed for one purpose, in one configuration, for controlled conditions. Drop them, break a joint, put them on gravel instead of a lab floor — and they stop working. Real life doesn't cooperate with fragile machines.
Northwestern University engineers have built something fundamentally different.
Legged Metamachines: Robots Made of Robots
Published in the Proceedings of the National Academy of Sciences on March 6, 2026, the research from Professor Sam Kriegman's lab introduces "legged metamachines" — modular robots made from autonomous, Lego-like units that combine into an endless variety of configurations.
Each individual module is itself a complete robot: its own motor, battery, onboard computer, and sensors. Alone, a module can roll, turn, and jump. But when multiple modules snap together, something remarkable happens: the combined machine takes on new movement styles and a kind of indestructibility that no single robot could possess.
Depending on their configuration, the metamachines undulate like seals, bound like lizards, or spring like kangaroos. They can flip themselves upright when overturned, hop over obstacles, and perform aerial acrobatics like spinning in mid-air.
AI-Evolved Body Plans
The team didn't design these configurations manually. Instead, they used an evolutionary algorithm — modelled on natural selection — to generate and test thousands of possible robot body designs in simulation. The AI churned out "species" of machines that no human engineer would have conceived, selecting for configurations that moved efficiently and survived damage.
"Instead of sticking with standard dog- or human-like designs, the AI found strange new forms," the researchers explained. The weirdest ones got built in real life — and they worked.
Refusing to Die
The most remarkable property of metamachines is their resilience. Because each module is itself an independent robot, a damaged metamachine doesn't simply stop. When a "leg" breaks off or a machine is deliberately chopped in half, the severed piece keeps functioning — and navigates back to rejoin the larger machine.
The team tested their creations on rugged real-world terrain: gravel, grass, mud. The machines kept going where rigid, purpose-built robots would have failed completely.
Why This Matters
The implications reach far beyond lab demonstrations. Robots that can reconfigure themselves and recover from damage could operate in disaster zones, collapsed buildings, or remote environments where repair is impossible. They could adapt to conditions they were never explicitly programmed for.
More broadly, the research demonstrates that AI-driven evolution can find solutions to physical problems that human engineers wouldn't reach through conventional design — a principle that could reshape how robots, and perhaps other machines, are built.
Nature has spent hundreds of millions of years evolving bodies that adapt and survive. For the first time, we have a glimpse of machines that might do the same.
Sources: Northwestern University McCormick School of Engineering (March 6, 2026) · Proceedings of the National Academy of Sciences · New Atlas · IFLScience · New Scientist
