A new frontier in robotic autonomy: metabolism

Introducing “robot metabolism”: machines that assemble, heal, and evolve themselves

Robots have long been built as tools—preassembled machines with a fixed purpose and a fixed shape. They operate within strict design boundaries, relying on human input for upgrades, repairs, and adaptation. But what if robots could change themselves? What if they could grow, heal, and evolve the way living organisms do—by reconfiguring their own bodies?

That’s exactly what researchers at Columbia University are working on. They’ve created robots that don’t just operate in the world; they interact with it physically, metabolically, and structurally. In doing so, they may have redefined what it means for a machine to be autonomous. Described in a new study published in Science Advances, this new process, called “Robot Metabolism,” enables machines to absorb and reuse parts from other robots or their surroundings.

From sticks to structures: how robot grow

At the heart of this innovation is a modular unit called a Truss Link—an extendable, stick-like element equipped with magnetic connectors that snap together without the need for precise positioning. A single link is capable of minimal movement, but when multiple links combine, the robot begins to evolve.

The assembly process unfolds in stages, beginning with simple two-dimensional shapes like triangles and stars, which then merge into more complex configurations such as a “diamond with a tail.” This intermediary shape folds itself into a tetrahedron—a structure that can move in three dimensions by toppling over. Finally, by attaching an additional link as a kind of limb, the tetrahedron becomes a “ratchet tetrahedron,” increasing its downhill speed by over 66%. With each step, the robot gains new mechanical capabilities, physical intelligence, and spatial freedom. This carefully designed transformation mimics biological growth: form follows function, and complexity emerges from simplicity.

Healing, replacing, evolving: biology in motion

The similarities to biological systems don’t stop at development—they extend to resilience: these robots are capable of self-repair. In test scenarios, when connections were severed by impact, the robots reformed their original shapes. If a module’s battery ran critically low, it detached automatically and was replaced by a functioning component, much like the biological process of apoptosis. What’s more, robots can assist each other: fully assembled structures have been used to lift and guide incomplete units into their final forms. The researchers refer to this as robot-assisted assembly, and it hints at a future where machines don’t just function independently—they collaborate, share resources, and build one another.

This emerging “robot ecology” may become essential in environments where human maintenance isn’t an option.

Adaptive machines for an unpredictable world

The potential applications are vast and compelling. In disaster zones or space exploration, where resources are limited and remote repair is impossible, robots capable of self-repair and physical adaptation could prove invaluable. These machines wouldn’t just survive harsh conditions—they’d reconfigure to meet them. Their physical autonomy would match their computational intelligence, allowing them to become long-term, self-sustaining systems. Robot metabolism may offer a digital interface to the physical world that enables true autonomy: machines that grow smarter and stronger with time, without external input. As lead author Philippe Martin Wyder explains, this creates an entirely new dimension of machine evolution—one that may be more useful than cognition alone.

Between innovation and responsibility

While the possibilities are exciting, they’re not without consequence. The idea of machines that grow by consuming parts—especially from one another—evokes familiar sci-fi tropes and raises ethical concerns. Autonomy in machines is no longer just about thought. If we expect robots to function without human support, they’ll need to be able to take care of themselves. That means growing, healing, adapting—and potentially evolving in ways we didn’t plan. The Columbia team isn’t just building robots; they’re laying the foundation for a new kind of artificial life, one governed less by programming and more by potential. As we enter a future filled with intelligent machines, the real question may no longer be how smart they are—but how alive.