Multi-Robot Collaborative Exploration on Communication-Constrained Environments

Abstract

Multi-robot cooperative exploration systems can gather environmental data in harsh and unknown scenarios, reducing the risk of human exposure and enhancing exploration capabilities. However, some communication-constrained (e.g., post-disaster) scenarios hinder the ability of robots to exchange map information. To resolve this issue, existing approaches utilize control strategies or deep reinforcement learning to enhance the map exchange process and the associated decision-making algorithm. However, the performance of these methods is limited in such scenarios due to the following challenges: 1) the extensive exchange of redundant information increases the burden on the robots’ communication bandwidth; 2) exploration capabilities are limited when map information transmission is blocked. To overcome these challenges, we propose a novel framework entitled CMICE (Communication-constrained Multi-Robot Intelligent Collaborative Exploration). Specifically, the framework contains the following two components: 1) a Non-Essential Non-Communication (NENC) component is introduced to reduce bandwidth requirements significantly; 2) a CNN-based Multi-agent TD3 (CMATD3) algorithm is designed to enhance exploration capabilities in the above scenarios. Experimental results demonstrate that CMICE outperforms the state-of-the-art models in both ideal and communication-constrained environments.