Learning to cooperate together: A semi-autonomous control architecture for multi-robot teams in urban search and rescue

The goal of cooperative rescue robot teams in urban search and rescue (USAR) missions is for the rescue robots to effectively work together in order to minimize the overall exploration time it takes to search disaster scenes and find as many victims as possible. To achieve this goal, task allocation and execution amongst the team members must be considered. In this paper, a unique hierarchical reinforcement learning (HRL) based semi-autonomous control architecture is proposed for rescue robot teams to enable cooperative learning between the robot team members. The HRL-based control architecture allows a multi-robot rescue team to collectively make decisions regarding which rescue tasks need to be carried out at a given time, and which team member should execute them to achieve optimal performance in exploration and victim identification. Due to the cluttered nature of disaster scenes, we propose the development of a semi-autonomous centralized control approach to allow task sharing between the robot team members and human operators when needed. Simulation results verify the effectiveness of the proposed HRL-based methodology for multi-robot cooperative exploration and victim identification in USAR-like scenes.