Multi-Robot Collision Avoidance with Map-based Deep Reinforcement Learning

Multi-robot collision avoidance in a communication-free environment is one of the key issues for mobile robotics and autonomous driving. In this paper, we propose a map-based deep reinforcement learning (DRL) approach for collision avoidance of multiple robots, where robots do not communicate with each other and only sense other robots' positions and the obstacles around them. We use the egocentric grid map of a robot to represent the environmental information around it, which can be easily generated by using multiple sensors or sensor fusion. The learned policy generated from the DRL model directly maps 3 frames of egocentric grid maps and the robot's relative local goal positions into low-level robot control commands. We first train a convolutional neural network for the navigation policy in a simulator of multiple mobile robots using proximal policy optimization (PPO). Then we deploy the trained model to real robots to perform collision avoidance in their navigation. We evaluate the approach with various scenarios both in the simulator and on three differential-drive mobile robots in the real world. Both qualitative and quantitative experiments show that our approach is efficient with a high success rate. The demonstration video can be found at https://youtu.be/jcLKlEXuFuk.