Neural-Adaptive Switching Control of Task-Space Objectives on Strict-Feedback Robot

Abstract

Switching between tasks in robot applications is necessary for autonomous behaviour, requiring a control framework for switching between objectives. Thus, this paper proposes a switching control framework that allows transitions in position-dependent objectives for robot systems in strict-feedback form with adaptive capabilities. The proposed controller integrates artificial neural networks in two ways; one approximates control transforms while another approximates additive terms. Furthermore, the proposed method includes a dynamic damped inverse to ensure the invertible control transform matrices. The proposed controller is derived using the backstepping method and is stable in a Lyapunov sense. The proposed task-switching controller is first demonstrated on a generic robot manipulator. Then, a simulated two-degree-of-freedom manipulator with a camera at the end-effector tasked with position tracking and visual servoing verifies the capabilities of the proposed controller. A baseline switching controller shows that the proposed controller has superior performance, specifically, root mean square (RMS) error and control effort. Also, executing the simulation for multiple cycles shows that the proposed controller's artificial neural network and adaptive laws are stable.