Cornell robots turn shape into intelligence

Cornell engineers have built a robotic collective that moves, reshapes and adapts without a central controller, marking a fresh advance in efforts to design machines whose behaviour emerges from physical interaction rather than software commands alone.

Called the Cross-Link Collective, the system uses dozens of small robotic modules that are weak and inefficient on their own but capable of sustained group motion when they connect, disconnect and reorganise. The work, published on May 20, points to a robotics model closer to flowing matter than to a conventional machine, with movement shaped by contact, friction, entanglement and simple local responses.

Each module is about 200 millimetres long and 20 millimetres wide. A small motor drives it to oscillate between two shapes, resembling an “I” and a “U”. That motion lets the modules inch forward and bump into one another. Weak Velcro patches at their ends allow them to latch temporarily, forming shifting chains that can pull, bend and release as conditions change.

“Instead of relying on explicit computation and communication, the system shifts the intelligence into the shape of the robots and their physical interactions,” said Kirstin Petersen, associate professor of electrical and computer engineering at Cornell and corresponding author of the study. “We’re leveraging the contact dynamics to let useful behaviours emerge, so the system naturally settles into configurations that reduce internal stresses and improve motion.”

The approach challenges a dominant assumption in robotics: that reliable behaviour requires precise sensing, computation and control. The Cornell design gives up exact command over every unit, but gains robustness through redundancy. If one module slows down, loses power or becomes poorly positioned, the collective can continue moving because no single unit determines the group’s function.

Tests showed that linked chains outperformed individual modules on inclined surfaces, where single robots often stalled depending on their orientation. In obstacle fields, the collective behaved more like a deformable material than a fixed mechanism, forming bonds to maintain cohesion and breaking them to avoid jamming. That capacity to flow around disruption is central to the system’s appeal.

Danna Ma, the study’s lead author and a visiting lecturer in electrical and computer engineering, said the collective remains functional because its structure is not dependent on perfect components. “It doesn’t matter if one module has a compromised battery or fails for other reasons,” Ma said. “The system stays functional because it can adapt. It is redundant and doesn’t depend on any single module.”

The researchers also showed that limited computation can improve performance without turning the system into a centrally managed swarm. Isolated modules emit an audible distress signal when they infer that they have lost contact with the group. Nearby modules then slow down, giving the straggler a chance to reconnect. The signal is simple, but it helps cohesion while preserving the principle of local interaction.

“There is no centralized sensing or control,” Ma said. “Each module can infer when it has lost contact with the group by how much it’s being jostled and then use an audible buzz to slow down nearby modules while it catches up. It’s as simple as that.”

The project grew from earlier work on modular and soft robotic systems, including designs developed with collaborators at the Georgia Institute of Technology. Petersen and Ma refined the modules over years of experimentation and statistical analysis, studying how size, connection strength and shape influence entanglement and collective movement.

The system draws inspiration from active gels, materials whose molecular links continually form and dissolve while retaining a broader structure. That analogy is important because the Cross-Link Collective does not simply imitate an animal swarm; it behaves as a form of robotic matter whose properties arise from interactions among many small units.

The broader robotics field is moving towards embodied intelligence, where physical form, actuation and environmental contact contribute to decision-making. Humanoid robots, warehouse systems and autonomous vehicles often depend on expensive sensing and powerful models. Cornell’s work highlights another route: designing bodies and connections so useful behaviour emerges naturally from mechanics.

The technology remains experimental. The modules are small, slow and designed for laboratory study rather than immediate deployment in factories, disaster zones or construction sites. Scaling the approach will require stronger materials, better energy systems and deeper understanding of how large collectives behave under complex loads.