Key points are not available for this paper at this time.
In Urban Search and Rescue (USAR) missions, mobile rescue robots need to search cluttered disaster environments in order to find victims. However, these environments can be very challenging due to the unknown rough terrain that the robots must be able to navigate. In this paper, we uniquely explore the first use of deep reinforcement learning (DRL) to address the robot navigation problem in such cluttered environments with unknown rough terrain. We have developed and trained a DRL network that uses raw sensory data from the robot's onboard sensors to determine a series of local navigation actions for a mobile robot to execute. The performance of our approach was successfully tested in several unique 3D simulated environments with varying sizes and levels of traversability.
Zhang et al. (Wed,) studied this question.