Consensus-based Deep Reinforcement Learning for Mobile Robot Mapless Navigation

Abstract

When using mobile robots to perform data collection about the surroundings, the performance might be dissatisfying since the environments could be unknown and challenging. This situation will pose challenges for mobile robot navigation and exploration. To tackle this issue, we propose a consensus-based deep reinforcement learning (DRL) algorithm for multiple robots to perform mapless navigation and exploration. The proposed algorithm leverages both consensus-based training and DRL, which reduces required training steps while maintaining the same training reward. Once trained with fixed obstacles, the proposed training model can demonstrate adaptability in handling real-world random static obstacles and sudden obstacles. The experimental video is available at: at: https://youtu.be/ym2yvbKg4fU.