Adaptive image-based visual servoing with reinforcement learning for wheeled mobile robots

Appropriate servoing gain are critical to good performance of image-based visual servoing (IBVS). Servoing gain affects the stability and the convergence rate for the robot to reach a desired position, but the servoing gains in many IBVS applications are heuristically set as a constant. A generic method for determining a series of the servoing gains is proposed, which adjusts adaptively the servoing gain by using Q-learning in order to realize more efficient control. The proposed method addresses problems associated with IBVS control, for instance, slow convergence and low stability. The complete IBVS control system is validated by several experiments on a WMRs that reaches a desired position. Simulation and experimental results demonstrate that the proposed IBVS method has better convergence and stability than the competing methods.