This paper aims to develop a novel model-free controller for uncertain robot manipulators that achieves high-accuracy tracking performance with adaptive feedback gains.
This study uses sparse online Gaussian processes (SOGP) to model unknown robot dynamics and design the motion tracking controller. SOGP facilitates online updates using a sensor data stream. The authors integrate SOGP into a robust sliding mode controller for trajectory tracking and design adaptive feedback gains proportional to the predicted variance from SOGP, balancing response speed and robustness.
Simulation results for a 2-degree of freedom robot manipulator indicate that the proposed SOGP-based controller outperforms GP, radial basis function neural network and PD controllers, particularly in unexplored regions of the state space. Furthermore, the proposed method maintains tracking accuracy under external perturbations, such as payload changes and disturbances.
SOGP facilitates online adaptation and enhances tracking in unexplored regions compared to offline-trained GP models. Adaptive feedback gains, based on SOGP confidence, balance response speed and robustness. The method demonstrates effectiveness through practical simulations involving payload changes and disturbances.
