Abstract: Aiming at the problem of decreased manipulator manipulability and high trajectory error caused by the Dynamic Movement Primitive (DMP) method,neglecting joint kinematic constraints during trajectory planning,a new method of Manipulability-based DMP (M-DMPs) was proposed.Through numerical solutions,the current manipulability gradient of the redundant manipulator was determined,and subsequently adjusted the external force coupling terms of DMPs separately based on the end-effector velocity component and null space component.Then,to ensure manipulability during the operation of the robotic manipulator,the dynamic parameters were designed to autonomously adapt and regulate the size of the external force coupling terms.To address the original method's obstacle avoidance limitations and potential manipulability reduction during obstacle avoidance,an exponential function optimized potential field was coupled with M-DMPs.The stability of the dynamic system was demonstrated through theoretical analysis.Simulations and physical experiments were conducted in various scenarios,demonstrating that the M-DMPs method reduced trajectory tracking errors by 53.56% while increasing manipulability by 27.69%,effectively avoiding obstacles.

Key words: imitation learning, dynamic movement primitives, manipulability, trajectory planning

摘要： 针对动态运动基元方法(DMP)在轨迹规划时忽略关节运动学约束,导致机械臂灵活度下降、轨迹跟踪精度低的问题,提出一种基于可操作性的改进版DMP(M-DMP)。该方法首先通过数值解确定冗余机械臂当前的可操作性梯度,并分别从末端速度分量和零空间分量调整动态运动基元的外力耦合项。然后设计了动态参数,能够自适应调节外力耦合项大小,保证机械臂运行时的灵活度。其次为解决原方法无法避障,以及避障过程中可能出现灵活度降低的问题,利用指数函数优化势力场后,将其与M-DMP结合实现避障。通过理论分析证明了M-DMP的稳定性。最后在多种场景下进行仿真和实物实验,结果表明M-DMP方法在避开障碍物的同时,降低了53.56%的轨迹跟踪误差,并将可操作性提高了27.69%。

关键词: 模仿学习, 动态运动基元, 可操作性, 轨迹规划