Predictive Planning Method for Rescue Robots in Buildings

Rescue robotics is a domain with great potential for important, real-life applications. Recently its development mostly focused on creating machines able of moving in collapsed buildings remains, leaving group operations in undamaged buildings slightly neglected. In this paper we are focusing on methods for managing multi-robot rescue actions in buildings, where some damages should be expected, but the general structure of a building remains unchanged. We propose a predictive planning approach, which allows using time demanding optimization algorithms for the task, where environment can change rapidly. The proposed solution predicts possible exceptional situations and pre-calculates alternative plans. A prototype system for parallel calculating of plans using high performance computing hardware is presented. A comparison of the optimized plans with a greedy approach is shown for different buildings and different sizes of robot teams. Finally the evaluation of the approach on real mobile robots is described.