The state of the art of robot learning control using artificial neural networks—an overview

Abstract

This article examines state-of-the-art learning control schemes, particularly in applications for robot control systems, based on artificial neural network architectures with supervised and unsupervised learning rules. Existing schemes are classified as belonging to two different approaches: inverse dynamics modeling robot learning control (IDMRLC) and sensorimotor coordination robot learning control (SMCRLC). First, some commonly used artificial neural network architectures and learning rules are briefly reviewed. The inverse dynamics modeling robot learning control approach and the sensorimotor coordination robot learning control approach are then presented. Finally, additional neural-based solution approaches for other robot control problems are discussed. Ongoing research is showing that learning control schemes based on artificial neural network architectures are evolving to become a powerful technique for building stable, adaptive, real-time, and robust controllers. In addition, in the meantime they can help validate more biologically inspired models.