Socially Compliant Navigation through Raw Depth Inputs with Generative\n Adversarial Imitation Learning

Abstract

We present an approach for mobile robots to learn to navigate in dynamic\nenvironments with pedestrians via raw depth inputs, in a socially compliant\nmanner. To achieve this, we adopt a generative adversarial imitation learning\n(GAIL) strategy, which improves upon a pre-trained behavior cloning policy. Our\napproach overcomes the disadvantages of previous methods, as they heavily\ndepend on the full knowledge of the location and velocity information of nearby\npedestrians, which not only requires specific sensors, but also the extraction\nof such state information from raw sensory input could consume much computation\ntime. In this paper, our proposed GAIL-based model performs directly on raw\ndepth inputs and plans in real-time. Experiments show that our GAIL-based\napproach greatly improves the safety and efficiency of the behavior of mobile\nrobots from pure behavior cloning. The real-world deployment also shows that\nour method is capable of guiding autonomous vehicles to navigate in a socially\ncompliant manner directly through raw depth inputs. In addition, we release a\nsimulation plugin for modeling pedestrian behaviors based on the social force\nmodel.\n