Dual-Layer Path Planning With Pose SLAM for Autonomous Exploration in GPS-Denied Environments
Zhang Shi, Rongxin Cui, Weisheng Yan, Yinglin Li
Abstract
Robot exploration in GPS-denied environments is a significant challenge due to the lack of reliable localization strategies. In this article, we propose a dual-layer planning approach with pose SLAM for autonomous robot exploration, which consists of local and global planners. The local planner uses the iteratively built local tree to generate candidate paths, and assesses the optimal path using a utility function that trades off exploration efficiency with localization accuracy. When the local planner is unable to return the admissible paths, the global planner is engaged to search an incrementally built graph for a path, which can reposition the robot to a previously identified valuable pose. For global path searching, we improve the Dijkstra algorithm and propose a cost function that considers localization uncertainty to generate a path with low localization error. In addition, we present a mixed filter on Lie group to estimate state information of paths for planners online. Finally, the proposed method is evaluated in challenging simulations and real-world environments. Comparison experiments show that our method is more efficient than the existing methods in exploring GPS-denied environments.