An information-driven POMDP framework for multi-robot formation in autonomous search and rescue

Efficient exploration in unknown environments is a critical challenge for search and rescue (SAR) robots. Existing methods neglect the balance of movement cost and rescue priority, rely on prior maps, and suffer from redundant paths and high communication overhead. This paper proposes an information-driven POMDP-based multi-robot formation framework for map-free autonomous SAR. We design a POMDP model with environmental semantics and human experience for dynamic cost-priority balance, a grid centroid-based topological path method for full coverage, and a distributed node fusion communication strategy for low-overhead collaboration. Simulation and real-world experiments show that our method reduces SAR completion time by up to 22.1%, cuts 90% of victim rescue time by 36.2%, and lowers path repetition by over 30% compared with state-of-the-art methods. It significantly improves the exploration and collaborative efficiency of multi-robot SAR systems, providing a robust solution for time-critical SAR missions.