ANN Control for tendon-driven robot mechanisms based on input-output strict linearization

摘要

We propose an adaptive neural network controller for tendon-driven mechanisms with nonlinear elastic elements. The controller is based on the input-output strict linearization of only robot dynamics, while the non-linear of the tendon is not compensated. To ensure the robustness, we use adaptive elements for the inertia matrix and tendon tensile force functions, and neural networks for the gravity force, coriolis-centrifugal forces and the motor dynamics. We proof a sufficient condition for the uniform ultimate boundedness of the trajectories, then show the efficiency by simulations.