To track a desired trajectory better, a CMAC neural network is added to conventional computed torque control as compensator. In the CMAC network, the inputs contain cross-coupling error composed of trajectory tracking error and synchronization error and its change rate, and the output is the compensation torque. Matlab simulations show that the tracking error and synchronization error can converge to zero with CMAC network. Compared with pure computed torque control, the trajectory tracking is more accurate with the proposed control strategy. The results show that the proposed control strategy is valid to track a predefined trajectory accurately. 