基于目标引导的多目标苹果采摘路径规划方法

Multi-objective Apple Picking Path Planning Method Based on Target Guidance

摘要:

针对苹果采摘机械臂在非结构化果园环境中路径规划效率低和路径质量差等问题，提出了一种结合树枝密集度参数的粒子群优化算法（Branch density parameter-based particle swarm optimization, BD-PSO）与目标引导快速扩展随机树星算法（Target-guided rapidly-exploring random tree star,TG-RRT*）的多目标路径规划方法（BD-PSO_TG-RRT*）。通过在快速扩展随机树星（RRT*）算法中引入自适应步长、设定等边圆锥采样区域、目标偏向策略和直连策略，加速路径生成并增强导向性。对初始路径进行冗余点去除与三次B样条曲线平滑处理，提升路径质量。通过BD-PSO算法确定多目标采摘顺序。实验结果表明，TG-RRT*算法相较于传统快速扩展随机树（RRT）和RRT*算法平均路径长度缩短23.18%、11.67%，平均时间降低12.59%、71.96%，平均迭代次数降低68.07%、31.58%。在多目标连续采摘路径规划仿真实验中，BD-PSO_TG-RRT*算法与原PSO与TG-RRT*结合算法相比，平均规划时间降低8.14%，平均迭代次数降低13.24%，BD-PSO_TG-RRT*算法能够生成适用于机械臂多目标采摘的最优路径，有效缩短了采摘路径总长度，并显著减少了路径规划时间。研究结果为苹果采摘机器人在执行多目标连续采摘任务时提供了技术参考。

Abstract:

Aiming to address the issues of low planning efficiency and long planning paths of apple-picking robotic arms in unstructured orchard environments, a target-guided multi-objective apple picking path planning method (BD-PSO_TG-RRT*) was proposed, which combined a particle swarm optimization (PSO) algorithm incorporating a branch density parameter with a target-guided rapidly-exploring random tree star (TG-RRT*) path planning algorithm. Firstly, based on the traditional RRT* algorithm, an adaptive step-size strategy was introduced, and an equilateral conical sampling region was defined. A target-biasing strategy was also incorporated to enhance the goal-directedness of sampling within this region. A direct connection strategy was used for new nodes to enable faster convergence, thereby improving the speed of path generation. Secondly, the initial planned path was refined by removing redundant points and transforming it into a smooth path using cubic B-spline curves, improving path quality. Lastly, to account for obstacles such as branches during the picking process, a branch density parameter was introduced into the PSO algorithm to obtain the optimal solution for the multi-objective picking sequence. Experimental results for path planning showed that compared with the RRT and RRT* algorithms, the TG-RRT* algorithm reduced average path length by 23.18% and 11.67%, respectively, decreased average time by 12.59% and 71.96%, and lowered the average number of iterations by 68.07% and 31.58%. In multi-objective picking experiments, the BD-PSO_TG-RRT* algorithm with the branch density parameter reduced average planning time by 8.14% and average iterations by 13.24% compared with the original PSO combined with TG-RRT* algorithm. These experimental results demonstrated that the BD-PSO_TG-RRT* algorithm accurately generated an optimal path for multi-objective applepicking, shortened the path length, reduced planning time, and significantly improved the efficiency of multi-objective apple-picking path planning. This algorithm can provide technical reference for apple picking robots to perform multi-objective continuous picking tasks.