An improved APF-guided BiRRT* motion planner for autonomous robotic apple harvesting in cluttered orchard environments

Abstract

Robotic apple harvesting in cluttered orchards is challenging because target fruits are often located in narrow spaces surrounded by branches, support wires, and other obstacles. In such obstacle-constrained environments, motion planning must not only generate collision-free trajectories but also ensure sufficient efficiency and execution reliability for harvesting operations. To address this problem, a six-degree-of-freedom (6-DOF) robotic arm apple-harvesting prototype was developed, and an improved APF-BiRRT* planner was proposed by integrating artificial potential field (APF) guidance into the bidirectional rapidly exploring random tree star (BiRRT*) framework. The method further enhances tree expansion through repulsion-enhanced guidance, line-biased sampling, density-aware goal biasing, adaptive step-size modulation, greedy path simplification, and local spline smoothing. In four representative simulated orchard harvesting scenarios, comparisons with seven selected planners showed that the proposed approach reduced planning time by 5.6–82.6%, shortened path length by 11.7–49.0%, and increased planning success rate by 2–52%. The planner was further implemented on the physical harvesting prototype to enable collision-free motion planning based on point-cloud perception. In real harvesting scenarios, the system achieved an average harvesting success rate of 90.0% with an average harvesting time of 7.2 s. These results demonstrate that the improved APF-BiRRT* planner enables efficient and reliable obstacle-aware motion planning for robotic apple harvesting in cluttered orchards, providing a practical foundation for autonomous harvesting systems in smart agriculture.