Belief consolidation in a partially observable multi-robot environment with limited data sharing capabilities

Abstract

Effective multi-robot search and rescue hinges on rapid exploration, robust situational awareness, and judicious use of scarce bandwidth. This paper introduces a hybrid coordination architecture that combines centralised information fusion with distributed on-board autonomy. At the heart of the approach is a lightweight server that maintains complementary short-term and long- term probabilistic maps of the workspace, ranks unexplored regions by information entropy, and assigns each robot a region whose expected utility maximises collective coverage. Robots operate under these high-level directives while retaining full local autonomy for motion planning and obstacle avoidance; they update the server only when new observations significantly alter the shared belief, thereby enforcing an event-triggered communication policy. The proposed framework advances the state of the art in three ways. First, the dual-layer map- ping strategy accommodates both transient sensor cues and persistent environmental structure without incurring prohibitive memory or update costs. Second, the entropy-driven region allocator dynamically balances exploration load and mitigates redundant traversal as environmental uncertainty evolves. Third, the selective communication rule maintains global consistency while sharply reducing channel utilisation, enabling scalability to larger robot teams and harsher communication conditions. A suite of synthetic disaster scenarios featuring dynamic obstacles and mobile victims is used to evaluate the architecture against fully centralised and fully decentralised baselines. Results demonstrate superior exploration efficiency, faster victim localisation, and lower communication cost, confirming that selective, uncertainty-aware information exchange strikes an effective balance between shared situational awareness and individual responsiveness. The findings suggest that hybrid coordination, underpinned by principled workload allocation and event-triggered messaging, offers a promising direction for time-critical multi-robot operations in complex, un- certain environments.