FORMICA: Decision-Focused Learning for Communication-Free Multi-Robot Task Allocation

A research team from Worcester Polytechnic Institute has introduced FORMICA (Field-Oriented Regret-Minimizing Implicit Coordination Algorithm), a breakthrough in multi-robot coordination that eliminates the need for communication between robots. The system addresses a critical limitation in swarm robotics: most existing methods degrade sharply in environments with limited bandwidth, degraded infrastructure, or adversarial interference where communication is unreliable or impossible.

FORMICA's key innovation is using decision-focused learning where robots coordinate implicitly by predicting teammates' bids on tasks. Instead of communicating to resolve conflicts, each robot anticipates competition for tasks and adjusts its choices accordingly. The approach employs a mean-field approximation that reduces computational complexity from O(NT) to O(T), making it scalable to large swarms. Inspired by Smart Predict-then-Optimize (SPO), the system trains predictors end-to-end to minimize Task Allocation Regret rather than traditional prediction error.

Experimental results demonstrate significant performance gains. In scenarios with 16 robots and 64 tasks, FORMICA improved system reward by 17% and approached the optimal Mixed Integer Linear Programming (MILP) solution. When scaled to massive deployments with 256 robots and 4096 tasks, the same model maintained a 7% performance improvement, showing strong generalization capabilities. Remarkably, training requires only 21 seconds on a standard laptop, enabling rapid adaptation to new environments. This represents a paradigm shift from communication-dependent coordination to predictive, implicit coordination that could transform applications in disaster response, military operations, and infrastructure inspection where communication channels are compromised.