Behavior-Aware Robot Navigation with Deep Reinforcement Learning

Robot navigation in a crowded area is difficult owing to the complexity of the natural world and the difficulties of simulating human interactions. In order for robots to safely and efficiently navigate near humans, it is imperative to include human behavior in robot navigation models. The prior techniques, on the other hand, isolate human behavior from models of robot navigation and solve robot navigation in a context-aware and partially observable setting. To us, this is the most rudimentary type of self-awareness that mobile robots may exhibit. We propose a risk-aware robot navigation model based on deep reinforcement learning that is capable of making improved navigational decisions. Our model provides safety, risk-awareness, and environmental understanding for mobile robots in both outdoor and indoor settings. We demonstrate our model's performance in both simulation and real-world robot environments.