EHSP: An Efficient Heuristic Sampling-Based Planner for Autonomous UAV Exploration

This article proposes an efficient heuristic sampling-based planner (EHSP) to improve the exploration intelligence and efficiency in unknown spaces with narrow openings. Different from existing sampling-based methods, EHSP uses the environment's heuristic prior knowledge to guide the sampling process of the planner. In detail, a boundary mixture features (BMFs) detection strategy is proposed to extract the heuristic points from point clouds of the light detection and ranging (LiDAR). Heuristic strategy based on heuristic points is used for guiding the rapid unmanned aerial vehicle (UAV) movement and the rapidly exploring random tree (RRT) sampling. First, an addressable history map (AHM), consisting of heuristic points and visited historical RRT nodes, is embedded in the global planner based on the k-dimension (k-d) tree for dead-end recovery and rapid exploration region migration. Second, to refinedly explore regions with narrow openings and reduce the backtracking costs, using the extracted heuristic points, EHSP includes a local planner with variable sampling parameters to avoid narrow entrance missing. Finally, a hybrid information gain is designed based on the Recursive Shadowcasting (RSC) algorithm to balance the local and global exploration. Simulation and experimental results show practical improvements of the proposed method in calculation time, exploration time, exploration completeness, path length, and generalization compared with the classic receding horizon 'next-best-view' planner (RH-NBVP) and Shadowcasting methods.