Multi-robot collaborative SLAM and scene reconstruction based on RGB-D camera

Abstract

Single-robot visual SLAM has problems such as slow mapping speed. This study proposed a multi-robot collaborative SLAM and scene reconstruction system based on an RGB-D camera. The system adopts a centralized structure. While several client robots collect RGB-D image data respectively and transmit the data to the server, the server runs a stand-alone visual SLAM system based on ORB-SLAM2 for each client, and stores the real-time mapping data in the map manager. At the same time, it detects whether the maps in the map managers meet the fusion conditions at a certain frame rate, and uses the map fusion algorithm to fuse when the conditions are met. In order to solve the problem that ORB-SLAM2 can only do semi-dense mapping, the system uses the 7-DoF poses of each client output by collaborative SLAM to reconstruct the dense map. We evaluated the performance of the system through simulations. Compared with the single-robot system, the collaborative SLAM in this system has a significant improvement in speed and can maintain high accuracy. We verified the effectiveness of the scene reconstruction algorithm through the scene reconstruction simulation and further proved that high-precision camera poses can be obtained in collaborative SLAM through the reconstruction results.