ADAPTIVE ROBOT CONTROL USING NEURAL NETWORKS

This paper studies the trajectory tracking problem to control the nonl inear dynamic model of a robot using neural networks. These controller s are based on learning from input-output measurements and not on para metric-model-based dynamics, Multilayer recurrent networks are used to estimate the dynamics of the system and the inverse dynamic model. Th e training is achieved using the backpropagation method. The minimizat ion of the quadratic error is computed by a variable step gradient met hod. Another multilayer recurrent neural network is added to estimate the joint accelerations. The control process is applied to a two degre e-of-freedom (DOF) SCARA robot using a DSP-based controller. Experimen tal results show the effectiveness of this approach. The tracking traj ectory errors are very small and torques expected at manipulator joint s are free of chattering.