A Harvard School of Engineering and Applied Sciences team finds that adding more robots to a crowded arena slows the group down, but injecting a small, tunable randomness into each robot's path breaks the gridlock.
In a Harvard SEAS benchtop experiment, dense robot swarms hit a throughput ceiling. Adding more bots made the crowd slower, not faster, until researchers gave each robot a small, tunable randomness in its path. That controlled noise broke local deadlocks the way thermal motion lets particles escape traps, lifting the density ceiling without any central coordination.
The result, reported today by a Harvard John A. Paulson School of Engineering and Applied Sciences team led by applied mathematician L. Mahadevan, is a clean demonstration of a design principle: in confined crowds, interference creates a jamming transition where extra units become a liability. The team's release, republished by SciTechDaily describes a wheeled-robot arena where throughput rose with swarm size up to an optimum, then collapsed as crowding and physical interference took over.
The lever is counter-intuitive. Each robot is treated as an agent following simple local rules, with a small, controlled wiggle injected into its motion. That stochasticity averages out the local deadlocks that form when deterministic paths collide and lock in place. The underlying physics is established in active-matter research. The contribution here is a swarm-robotics demonstration in a specific density regime, not a claim of new physics.
Lead author Lucy Liu, an applied mathematics Ph.D. student co-advised by Justin Werfel, frames the operating range directly: below the interference threshold, randomness adds nothing. Above it, randomness is what makes averages possible. The motivating use cases, including oil spill cleanup and complex equipment assembly in tight spaces, appear in the release as targets rather than as demonstrated outcomes. A figure credited to Lucy Liu and Harvard SEAS shows the benchtop arena, not a deployed swarm.
The work sits in a broader tradition of treating collective behavior as an emergent property of simple local rules. Pedestrian crowd dynamics and granular flow both show jamming transitions of the same mathematical shape, and the researchers invoke the analogy without claiming a deployment-ready pedestrian model. Translating the principle to a real-world swarm will require matching the noise injection to the actual interference pattern of the target environment, not assuming that more coordination will fix what a small amount of randomness can break.