Data-Driven Kinematic Control with Adaptive Gains for a Redundant Robot

This paper proposes a control law with adaptive gains based on a data-driven kinematic model for a redundant robot, which is considered as an unknown, nonlinear and discrete-time Multi-Input Multi-Output (MIMO) system. The control law is tested for a trajectory tracking task of the end-effector. The Pseudo Jacobian Matrix computes an online equivalent kinematic model using the input and output signals. A neuro-fuzzy network structure is used to adapt the control gains using the updated Jacobian matrix for each end-effector's axis. The simulation results demonstrated a good performance of the control scheme considering the robot trajectory with and without disturbances during the task execution and a considerable reduction of the estimation model error.