Multi-robot belief propagation for distributed robot allocation

Establishing collaborative behavior is an important factor in coordinating teams of robots. Multi-robot task allocation is one aspect of group coordination that deals with the assignment of robots to subtasks. Toward this end, we propose Multi-Robot Belief Propagation (MRBP), a synthesis of distributed probabilistic inference and notions of theory-of-mind for the purpose of multi-robot task allocation. MRBP does not rely upon a central planner or fixed decision hierarchy, but allows to infer task assignments locally through Bayesian belief propagation. The belief propagation algorithm provides us with a tool to incorporate task-related intentions and beliefs of robots into the assignment process. An assignment process in which each individual robot probabilistically decides, based on its own observations and the intentions and beliefs of other robots, which task it should allocate itself to. As a result, MRBP provides a means for collaborative group behavior without explicit protocols and command hierarchies. We present a sample implementation of MRBP for the chain-of-sight task allocation problem and show the results obtained from physically simulated robot teams.