Dynamics parameter identification of robots via experiment is the main way to obtain parameters of modelbased motion controllers. Traditional identification methods can only identify linear dynamics models, therefor their identification accuracies are limited. In order to solve this problem, parameter identification using artificialbeecolony(ABC) algorithm is proposed. The dynamics model of an industrial robot is established using NewtonEuler method. A nonlinear friction model which performs better in low speed dynamic characteristic is used to describe joint friction. Then excitation trajectory is optimized. After necessary preprocess of experimental data is done, dynamics parameters of the robot are obtained using ABC algorithm. As the result, dynamics parameters can be effectively obtained using ABC algorithm. Error peaks on predictive torque curves are also restrained by compensating the model. On this basis, the modelbased feed forward controller is designed. The tracing accuracy is remarkably improved with the help of feed forward controller.