Abstract: To solve the problems of low search efficiency and poor adaptability of the asymptotically optimal Rapidly-exploring Random Tree (RRT*) algorithm incorporating gravitational function in unstructured environment,and further improve the autonomous motion planning capability of manipulator,an adaptive improved RRT* algorithm based on Gaussian Potential Field (GPF) named GPF-RRT* algorithm was proposed.The unstructured environment perception was accomplished using a state-aware network.The optimized sampling point selection strategy was used to increase the sampling efficiency,and the GPF was used to construct the force potential field to improve the node search direction and step length for enhancing the algorithm's obstacle avoidance path planning capability in complex environments,which completed the pose optimization of the manipulator by combining with the manipulability and minimum energy index.Experiments based on the GPF-RRT* algorithm were conducted in different environments to compare with the P-RRT* algorithm in three aspects:search efficiency,path length and stability.The results showed that the proposed algorithm improved the average search efficiency by 38.06%,shortened the average path length by 46.6mm,which had strong stability in a variety of environments,and could effectively avoid the local minima problem.In addition,the effectiveness of the algorithm was further verified by the autonomous grasping operation of the manipulator.

Key words: manipulator motion planning, unstructured environment perception, asymptotically optimal rapidly-exploring random tree, manipulability and minimum energy

摘要： 为解决融合引力函数的渐进最优快速扩展随机树(P-RRT*)算法在非结构化环境下搜索效率低、适应性差等问题,进一步提高机械臂自主运动规划能力,提出一种基于高斯势场(GPF)自适应的改进RRT*算法(GPF-RRT*)。首先,利用状态感知网络完成非结构化环境感知。其次,利用优化后的采样点选取策略提高采样效率,通过GPF构建力势场改进节点搜索方向及步长,进而提高算法在复杂环境下的避障路径规划能力,并结合可操作度和最小能量指标完成机械臂的姿态优化。最后,基于GPF-RRT*算法在不同环境下进行实验,与P-RRT*算法在搜索效率、路径长度、稳定性3个方面进行对比。结果表明,所提算法的平均搜索效率提高了38.06%,平均路径长度缩短了46.6 mm,在多种环境下均具有较强稳定性,且能有效避免局部极小问题。另外,通过机械臂自主抓取操作进一步验证了算法的有效性。

关键词: 机械臂运动规划, 非结构化环境感知, 渐进最优快速扩展随机树, 可操作度和最小能量