An Efficient Sampling-Based Planner towards Optimal Path Planning in Complex Environments

Efficient and robust path planning is crucial to ensure the planning efficiency and operational safety for autonomous inspection robot. While sampling-based methods are widely used in robot path planning due to their computational efficiency, these methods still face following challenges in real applications: (1) sub-optimal path due to slow convergence and (2) inherent trade-off between search efficiency and path quality. These limitations restrict their effectiveness in complex scenarios, where reliable navigation is required. To overcome these limitations, we propose the Swift Optimization Bidirectional RRT* (SOB-RRT*) that refines the balance between path optimization and planning efficiency in complex environments. First, we propose a foresighted path generation method that incrementally expands the optimization region of tree nodes based on their hierarchical connection, significantly reducing zigzags in the path and effectively mitigating local optima. Second, a flexible greedy connection strategy is proposed to accelerate the path search by dynamically selecting the connection target. Finally, we introduce a constrained swapping strategy to maintain the growth balance of two search trees. This guarantees sufficient optimization for paths in both two trees and prevents the planner from being trapped in complex scenarios by restricting swapping frequency. Simulations and real-world experiments show improvements of the proposed planner in terms of time cost, search efficiency, and navigation success rate.