Neuro-inspired reward-based tracking control for robotic manipulators with unknown dynamics

Tracking control for robotic manipulators is required for numerous automation tasks in manufacturing engineering. For this purpose, model-free PD-controllers are largely implemented by default in commercially available robot arms and provide satisfactory performance for simple path following applications. Ever more complex automation tasks however ask for novel intelligent and adaptive tracking control strategies. In surface finishing processes, discontinuous freeform paths as well as changing constraints between the robotic end-effector and its surrounding environment affect the tracking control by undermining the stable system performance. The lacking knowledge of industrial robot dynamic parameters presents an additional challenge for the tracking control algorithms. In this paper the control problem of robotic manipulators with unknown dynamics and varying constraints is addressed. A robust sliding mode controller is combined with an RBF (Radial Basis Function) Neural Network-estimator and an intelligent, biomimetic BELBIC (Brain Emotional Learning-Based Intelligent Control) term to approximate the nonlinear robot dynamics function and achieve a robust and adaptive tracking performance.