基于强化学习和意图推理的目标跟踪.pdfVIP

Proceedings of the 33rd Chinese Control Conference July 28-30, 2014, Nanj ing, China , , , 200240 E-mail: {lijiunderstand, jbsu}@sjtu.edu.cn : ,. , , . ,,, ; ,, ,.NAO .. : ; ; ; ; NAO Tracking Strategy Based on Reinforcement Learning and Intention Inference LI Jie, SU Jianbo School of Electronic Information and Electric Engineering , Shanghai Jiao Tong University, Shanghai 200240, P. R. China E-mail: {lijiunderstand, jbsu}@sjtu.edu.cn Abstract: Due to hysteresis in the visual information processing, the timeliness and performance is affected in the specific target tracking problem. This paper investigates a novel approach, target motion modeling through the grey prediction model, to compensate the uncertainty with the correction action of the robot. Firstly, with consideration of the nonlinear and non- gaussian characteristics of the environment and target tracking, the particle filter algorithm is introduced for motion prediction and estimation,and the target position in the robot vision can be calculated. Then, an intention inference based reinforcement learning control method is used to estimate the mapping from sensory information to appropriate robot action based on the characteristics of the method,which, do not need the model of the environment and the robot.At last, the method is used in humanoid platform NAO to realize robust people tracking problem. Experimental results show the validity of the proposed method. Key Words: grey prediction; particle filter; reinforcement learning; intention inference; NAO 1 , . , ,.