Multi-robot Exploration System in Unknown Environment Based on Submap

Abstract

When exploring the unknown environment, the efficiency of a single robot is too slow, while the map merging between robots is a difficulty when the initial pose of multiple robots is unknown. In order to solve these problems, we design a multi-robot exploration system in unknown environment based on submap, in which we design a robot map merging framework through the NDT-transformer network, The environment slam is implemented by Cartography, and the submap are transformed into 2D point clouds, which are matched and geometrically transformed by 2D- NDT-transformer to realize the map merging between robots, The average recall rate of the top 1% of the network is 7.75$\sim$ 11.77% higher than that of 2D-NetVLAD, and 12.44% $\sim$ 15.39% higher than that of 2D-PointNetVLAD. At the same time, in the aspect of path planning, we use a Multi-robot Potential Field RRT (MPRRT) exploration method, which combines RRT and artificial potential field, so that the exploration efficiency is greatly increased. Compared with ordinary RRT, the trajectory length is reduced by 4.9% $\sim$ 14%, the exploration time is reduced by 12% $\sim$ 17%, and the exploration efficiency is significantly improved.