Despite the fact that the inverse kinematics of 6R robot with closed solution can be solved analytically and numerically, it requires a huge amount of calculation. In addition, as for the inverse kinematics of manipulator, the classical particle swarm optimization (PSO) algorithm is characterized with such flaws as instability, local optimization and single population in many simulation experiments. Therefore, an improved PSO algorithm has thus been proposed; by introducing the dynamic weight factor, the dynamic weight adjustment factor is to be combined with CMA-ES algorithm so as to balance global search and local search ability; the contraction learning factor is introduced for an avoidance of falling into local optimum in the iterative process. Taking REBot-V-6R robot as an example, the forward kinematics model of the robot is established, and the inverse kinematics problem of the robot is transformed into the optimization problem of the improved PSO algorithm, with the position error and attitude error of the robot simulated respectively as well. Based on a comparison between the simulation results and those of the classical PSO algorithm and the genetic algorithm, it can be found that the performance of the proposed improved algorithm has been significantly improved in terms of solution accuracy and stability, thus verifying the feasibility and effectiveness of the algorithm.