Experiments in feedforward shaping control of direct-drive robot using RBF network

This paper considers a new approach to manipulator trajectory tracking. Feedforward control shape is computed with a radial basis function (RBF) network as a result of approximation over task parameter domain. The task parameters comprise coordinates of the initial and final point of the trajectory with motion time being fixed. The RBF network is trained using the feedforward shapes obtained with a learning control algorithm. The paper demonstrates experimental implementation of all steps of the algorithm for trajectory tracking in fast (1.25 s) motions of a direct-drive industrial robot AdeptOne. High performance is achieved in experiment opening an avenue for practical applications of the approach