Abstract: In order to improve the accuracy and robustness of dynamic parameters identification for manipulator, a parameter-identification method based on robust particle swarm optimization algorithm was proposed, taking into account both gross and random errors. A dynamic model of the manipulator was established, and the minimum dynamic parameter set was selected to eliminate parameter redundancy. In order to achieve sufficient parameter excitation, a finite Fourier series was used to construct the excitation trajectory, and the trajectory was optimized with the goal of minimizing the matrix condition-number. To improve property of parameter-identification, a new way of learning from other individuals was introduced into particle swarm algorithm; The addition of gross error detection function in the design of fitness function effectively improves the algorithm’s anti-interference ability. Through verification by manipulator system noising-experiment, the root mean square of the driving torque residual optimized by robust particle swarm optimization algorithm is 1.1975 N·m, which is 75.29% less than that of quantum particle swarm optimization algorithm; The correlation between the actual torque value and the calculated value is 0.9593, which is 11.73% higher than the quantum particle swarm algorithm. The experiment results show that the robust particle swarm optimization algorithm has higher accuracy and stronger robustness in the identification of manipulator dynamic parameters.