Neural computation of the equivalent control in sliding mode for robot trajectory control applications

In the application of sliding mode controllers, the main problem encountered is that a whole knowledge of the system dynamics (or inverse dynamics) and the system parameters is required to be able to compute the equivalent control. This is actually very rare in practice. In this paper, a feedforward neural network is proposed to compute the equivalent control. The weights of the net are updated such that the additional control term of the sliding mode converges to zero. Experimental studies carried out on a direct drive robot arm indicate that the proposed approach is a good candidate for trajectory control applications.