Towards autonomous premium tea harvesting: A high-efficiency and high-quality path planning based on optimized IABFMT* algorithm in unstructured canopies

Abstract

Autonomous robotic harvesting of premium tea remains a challenging task due to the complex and unstructured nature of tea canopies, where traditional path planning algorithms often struggle to balance computational efficiency with effective obstacle avoidance. To address this challenge, this paper proposes an optimized variant of the Informed Anytime Bi-directional Fast Marching Tree (IABFMT*) algorithm, named H-IABFMT*, specifically tailored for tea-plucking robots. The method integrates the Halton sampling strategy with a heuristic node expansion mechanism to improve both path quality and planning efficiency, while operating within a simulated premium tea harvesting environment that features octree-based dynamic obstacle mapping with probabilistic occupancy updates. Compared to IABFMT*, Informed RRT*, and BIT*, H-IABFMT* demonstrates superior performance in terms of algorithmic efficiency, solution quality, success rate, and stability across complex 2D and 3D environments. Furthermore, the simulation harvesting experiment shows that H-IABFMT* can rapidly generate high-quality initial paths in realistic tea harvesting scenarios and converge to near-optimal solutions within a limited time, indicating its strong potential for autonomous premium tea harvesting. This work establishes a solid foundation for future integration with physical robotic platforms and real-world field validation.