Autonomous Multirobot Navigation and Cooperative Mapping in Partially Unknown Environments

Abstract

To perform collaborative exploration tasks in outdoor environments, multi-robot systems require effective task planning and high-precision co-localization. However, there are many challenges in real-world environments, such as unavailable navigation maps and unpredictable obstacles. In this paper, we present a system architecture for autonomous multi-robot navigation and cooperative Simultaneous Localization and Mapping (SLAM). To enable accurate and efficient multi-agent navigation, this work proposes a hierarchical multi-robot path planning pipeline involving two tasks, i.e., global multi-robot planning and local navigation. The first task is formulated into a min-max k-Chinese postman problem and solved by a genetic algorithm. The second task is transformed into the autonomous collision-free movement of each robot and solved by a reinforcement learning method. To improve the flexibility of cooperative mapping in unknown outdoor environments, this work proposes an online multi-robot light detection and ranging (LiDAR) SLAM system, which can flexibly select heterogeneous robots equipped with different sensor combinations. Under the condition of no prior navigation map, this work realizes multi-robot cooperative environmental exploration and reconstruction. The proposed system architecture is tested and validated via real-world experiments and some public datasets. Experimental results exhibit the superior performance of the proposed method concerning accuracy, stability, and data efficiency.