Socially-compliant Navigation through Raw Depth Inputs with Generative Adversarial Imitation Learning

We present an approach for mobile robots to learn to navigate in pedestrian-rich environments via raw depth inputs, in a social-compliant manner. To achieve this, we adopt a generative adversarial imitation learning (GAIL) strategy for motion planning, which improves upon a supervised policy model pre-trained via behavior cloning. Our approach overcomes the disadvantages of previous methods, as they heavily depend on the full knowledge of the location and velocity information of nearby pedestrians, which not only requires specific sensors but also consumes much computation time for extracting such state information from raw sensor input. In this paper, our proposed GAIL-based model performs directly on raw depth inputs and plans in real-time. Experiments show that our GAIL-based approach greatly improves the behavior of mobile robots from pure behavior cloning both safely and efficiently. Real-world implementation also shows that our method is capable of guiding autonomous vehicles to navigate in a social-compliant manner directly through raw depth inputs.