In this paper, a new trajectory control approach of robot manipulator based on dynamics model was proposed. The method was a combination of the torque control and a compensator of neural network. The torque profile was updated by the previous torque profile. Errors were inevitable while the modeling for dynamics of robot manipulator, but they could be compensated by the compensation structure of ANN. As such, the trajectory tracking converged very fast and the new algorithm achieved the asymptotical convergence based on the Lyapunov’s method. Furthermore, it could enhance the performance of the adaptability. The result of simulations and experiments show that the controller can get the performance of trajectory well.