Force control for flexible robots using neural networks

Force control for flexible link robots using neural networks is considered. The nonlinear dynamics of the robot manipulator are identified through a recurrent neural network (RNN), which is trained in an off-line manner. Inversion of the RNN-based model dynamics leads to a feedforward component. The feedback controller gains are derived from the minimization of a discrete linear quadratic cost functional, subject to the model dynamics inferred by the linearization of the neural network along the desired trajectory. Sufficient conditions for temporal gain switching bounds are provided. The proposed control scheme is employed in simulation studies on a two link rigid-flexible manipulator