Indoor Multi-Sensory Self-Supervised Autonomous Mobile Robotic Navigation

Autonomous robotic navigation in indoor environments is fairly challenging and important to industrial environments. Traditional map-based or mapless navigation methods often fail because of the unstructured characteristics of the environments. Recently, imitation learning using DAgger algorithm has been successfully applied to many real-world robotic tasks. However, it needs human operators to give correct control commands without feedback to overcome data distribution mismatch problem, which is always prone to error and expensive. In this paper, we propose a novel solution to eliminate the need of human manual labeling after the initial data collection in the task of imitating to navigate in indoor environments. This solution introduces an imperfect policy based on multi-sensor fusion and a recording policy that only records the data giving the most knowledge to the navigation policy. The recording policy mitigates the affect of learning too much from an imperfect policy. With extensive experiments in indoor environments, we demonstrate that after several iterations of learning, the robot is able to navigate through real-world hallways in both seen and unseen situations safely. In addition, we show that our system achieves near human performance in most of the tasks and even surpasses human performance in one out of three tasks. To the best of our knowledge, this is the first work that utilizes imperfect sensor measurements to perform self-supervised imitation learning in robotic navigation tasks.