基于RL-DE改进算法的移动机械臂逆运动学求解研究OA

To address the inverse kinematics challenge arising from the joint redundancy in mobile platform-arm coupled systems,the traditional pseudo-inverse matrix-based methods are sensitive to singularity.This paper proposes an improved algorithm that integrates reinforcement learning(RL)and differential evolution(DE)to solve the inverse kinematics problem for highly redundant mobile robotic arms.The optimization model in the DE algorithm is constructed based on the physical model of the mobile robotic arms,with the fitness function defined as the error between the target pose and the desired end-effector pose of the mobile robotic arms.To overcome the limitations of fixed evolutionary parameters in conventional DE algorithm and enhance both global search capability and convergence speed,a dynamic mutation multi-strategy mechanism that adaptively adjusts evolutionary directions at different stages to balance global exploration and local exploitation is proposed.Furthermore,the deep deterministic policy gradient(DDPG)algorithm is employed to optimize the evolutionary parameters,in which variation in population fitness and diversity serves as state inputs,and the improvement rate of the best individual's fitness acts as the reward signal.Simulation results demonstrate that the proposed method eliminates the need for matrix inversion,effectively circumventing singularity issues and outperforming artificial bee colony,particle swarm optimization,and self-adaptive differential evolution algorithms in terms of inverse kinematics success rate and positioning accuracy.Additionally,it outperforms the self-adaptive differential evolution algorithm in trajectory smoothness and tracking precision,and exhibits lower average error in trajectory tracking.