Attention-Based Higher-Order Reasoning for Implicit Coordination of Multi-Robot Systems

Abstract

This paper presents a novel theory of mind (ToM)-based approach for implicit coordination of multi robot systems (MRS) in environments where direct communication is unavailable. The proposed approach integrates higher-order reasoning, epistemic theory, and active inference to coordinate the actions of each robot to clarify their own intentions and make them understandable to other robots. Further, to reduce the computational overhead of higher-order reasoning, we implement a large language model (LLM)-based attention selection mechanism that focuses on a subset of robots. Simulations and physical experiments demonstrate the applicability of the proposed approach with high success rates while significantly reducing computation complexity.