基于深度强化学习的机器人运动状态估计研究  

Research on robot motion state estimation based on deep reinforcement learning

作　　者：郑泛舟[1] Zheng Fanzhou(Department of Information Engineering,Meizhouwan Vocational Technology College,Putian 351119,China)

机构地区：[1]湄洲湾职业技术学院信息工程系,福建莆田351119

出　　处：《湖南文理学院学报（自然科学版）》2023年第1期34-39,共6页Journal of Hunan University of Arts and Science(Science and Technology)

基　　金：福建省教育厅基金项目(JAT201342)。

摘　　要：为了提升机器人运动状态估计的准确性与效率,所以研究基于深度强化学习的机器人运动状态估计方法。采用基于FPGA与DSP的图像采集技术采集机器人图像,利用彩色图像灰度化算法预处理采集到的彩色运动机器人图像,将灰度化处理后图像输入至深度强化学习网络,依次执行特征提取、搜索路径确定以及候选区域回归等操作,获得运动机器人目标检测结果。以机器人目标检测结果为主要依据,利用卡尔曼滤波算法完成机器人运动状态估计。实验结果表明,该方法可有效估计机器人运动状态,并且估计的准确性与效率均较高,受噪声环境影响较小,可以实现对于机器人运动状态的快速精准评估。In order to improve the accuracy and efficiency of robot motion state estimation,a robot motion state estimation method based on deep reinforcement learning is studied.The image acquisition technology based on FPGA and DSP is used to collect the robot image.The collected color moving robot image is preprocessed by the color image graying algorithm.The grayed image is input into the deep reinforcement learning network,and the feature extraction,search path determination and candidate region regression are performed in turn to obtain the moving robot target detection results.Based on the results of robot target detection,Kalman filter algorithm is used to estimate the robot motion state.The experimental results show that this method can effectively estimate the robot motion state,and the accuracy and efficiency of the estimation are high.It is less affected by the noise environment,and can realize the rapid and accurate evaluation of the robot motion state.

关 键 词：深度强化学习 机器人 运动状态估计 图像采集 图像灰度化 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]