| Visual target tracking task require the robot to accurately detect the target and finish tracking control based on visual feedback in real time in a dynamic environment.This task combines computer vision and robotic control,and presents a great challenge to target detection algorithm and tracking control algorithm.In this paper,a mobile robot control system,integrating visual module and controller module,is designed for visual target tracking task.The visual module performs target detection and feeds back error information to the controller module,which controls the robot to track the target.The visual module uses deep learning-based target detection algorithm,which has fast detection speed and high detection accuracy,and is very suitable for real-time target tracking.The controller is a combined neural network,RECMAC(Recurrent Emotional Cerebellar Model Articulation Controller),which is based on the mechanism of brain emotion learning with srtronger nonlinear approximation ability.In particular,the proposed network integrates CMAC a recurrent loop,which allows the network to remember the past states of the system and to learn knowledge of the system dynamics implicitly.The RECMAC network and a robust controller jointly form the controller module.The RECMAC network,acting as a primary controller,is designed for imitating an ideal controller,while the robust controller,performing as am indirect controller,is served for reducing the approximation errors between the ideal controller and the RECMAC.The Lyapunov stability theory is used to guarantee the stability of the global control system and derive the update laws of the RECMAC.The proposed system was validated and evaluated by both simulation and a practical moving-target tracking task.The experimentation demonstrated that the proposed system outperforms other popular neural network-based cont:rol systems,and thus it is superior in approximat:ing highly nonlinear dynamics in controlling vision-based mobile robot. |
PDF Full Text Request