MR-GMMExplore: Multi-Robot Exploration System in Unknown Environments based on Gaussian Mixture Model.

Yichun Wu,Qiuyi Gu,Jincheng Yu,Guangjun Ge,Jian Wang,Qingmin Liao,Chun Zhang,Yu Wang

DOI: https://doi.org/10.1109/ROBIO55434.2022.10011789

2022-01-01

Abstract:Collaborative exploration in an unknown environ-ment is an essential task for mobile robotic systems. Without external positioning, multi-robot mapping methods have relied on the transfer of place descriptors and sensor data for relative pose estimation, which is not feasible in communication-limited environments. In addition, existing frontier-based exploration strategies are mostly designed for occupancy grid maps, thus failing to use surface information of obstacles in complex three-dimensional scenes. To address these limitations, we utilize Gaussian Mixture Model (GMM) as the map form for both mapping and exploration. We extend our previous mapping work to exploration setting by introducing MR-GMMExplore, a Multi-Robot GMM-based Exploration system in which robots transfer GMM submaps to reduce data transmission and perform exploration directly using the generated GMM map. Specifically, we propose a GMM spatial information extraction strategy that efficiently extracts obstacle probability information from GMM submaps. Then we present a goal selection method that allows robots to explore different areas, and a GMM-based local planner that realizes local planning using GMM maps instead of converting them into grid maps. Simulation results show that the transmission of GMM submaps reduces approximately 96% communication load compared with point clouds and our mean-based extraction strategy is 4 times faster than the traversal-based one. We also conduct comparative experiments to demonstrate the effectiveness of our approach in reducing backtracking paths and enhancing cooperation. MR-GMMExplore is published as an open-source ROS package at https://github.com/efc-robot/gmm_explore.