Accurate and Efficient Multi-robot Collaborative Stereo SLAM for Mars Exploration

• Published in: 2022 International Conference on Service Robotics (ICoSR) pp. 118 - 125
• Main Authors: Shao, Yuanbin, Shao, Yadong, Wan, Xue
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.06.2022
• Subjects: Collaboration; Edge Computing Device; Location awareness; Mars; Mars Exploration; Multi-robot Collaboration; Performance evaluation; Real-time systems; Robot vision systems; Simultaneous localization and mapping; Stereo SLAM
• DOI: 10.1109/ICoSR57188.2022.00031

In recent years, planetary exploration has received a lot of attention in the aerospace field, and Mars is favored because of its cosmic environment that is very similar to the Earth. So far, human have sent six rovers and a helicopter to Mars. However, the GNSS global navigation system is unavailable on Mars, and there is a communication delay of 7 to 45 minutes between the Earth and Mars, which poses a huge challenge to the autonomous navigation and obstacle avoidance of the Mars robot. At the same time, the current exploration is carried out by a single robot, so the exploration range is limited. Multi-robot collaboration can improve the efficiency and robustness of planetary task execution. Multi-robot collaborative Simultaneous Localization and Mapping (SLAM) is conducive to enhancing the localization and mapping capabilities of robots. To achieve the goal, we propose an accurate and efficient Multi-robot collaborative stereo SLAM(MCS-SLAM). While ensuring that each robot works independently, MCS-SLAM collects the robot's localization and mapping results to the server through wireless communication, and completes the fusion optimization of multi-robot's localization and mapping data on the server. We generated six sets of image data, which were respectively captured by the stereo cameras carried by the simulated three rovers and three UAVs. Considering the limited CPU performance of Mars robot's computing device, we conducted experiments on Nvidia's edge computing equipment. The experimental results show that MCS-SLAM achieves real-time localization effects of 6fps and 10fps on Jeston TX2 and Jeston Xavier. Overall, when only stereo cameras are configured for collaborative work, the localization accuracy of the rover team and the UAV team reached 1.97m and 0.89m, respectively, and the average localization accuracy of 100 meters was 0.36m and 0.17m.