Socially Compliant Robot Navigation in Crowded Environment by Human Behavior Resemblance Using Deep Reinforcement Learning

Abstract

Social robots have evolved in diverse applications with the emergence of deep reinforcement learning methods. However, safe and secure navigation of social robots in a complex crowded environment remains a challenging task. The robot can safely navigate in a crowd only if it can predict the next action of humans, however this task becomes difficult because of the unpredictable human behavior. To address the issue of socially compliant navigation, the robot needs to learn real-time human behavior. This manuscript models Danger-Zone for the robot by considering all possible actions that humans can take at given time. The Danger Zones are formulated by considering the real time human behavior. The robot is trained to avoid these danger zones for safe and secure navigation. The proposed model is tested on the three state of art methods, Collision Avoidance with Deep Reinforcement Learning (CADRL), Long Short Term Memory Reinforcement Learning (LSTM-RL) and Social Attention with Reinforcement Learning (SARL) in multi-agent navigation. Experimental results signify that proposed model can understand human behavior and navigate in a socially compliant manner with safety as the highest priority.