A hierarchical reinforcement learning based control architecture for semi-autonomous rescue robots in cluttered environments

Teleoperated rescue robots designed to explore disaster scenes and find victims face serious limitations due to the cluttered nature of the environments as well as the rescue operators becoming stressed and disoriented in these scenes. An alternative to using teleoperated control is to develop fully autonomous controllers for rescue robots. However, these robots are also not capable of traversing these complex unpredictable environments. In order to address the limitations of both teleoperation and fully autonomous robotic control for urban search and rescue (USAR) environments, semi-autonomous controllers can be developed to allow task sharing and cooperation between a human operator and a robot. In this paper, a unique Hierarchical Reinforcement Learning (HRL) based semi-autonomous control architecture is proposed. The architecture provides the robot with the ability to learn and make decisions regarding which rescue tasks, exploration or victim identification, should be carried out at a given time and whether an autonomous robot or a human controlled robot can perform these tasks more quickly and efficiently without compromising the safety of the victims, rescue workers and the rescue robot. Preliminary experiments presented here evaluate the performance of the proposed HRL control approach for a rescue robot in an unknown cluttered USAR environment.