初始权值优化技术在机器人学习中的应用  

Implementation of Optimal Weights Initialization Technology in Robot Learning

作　　者：肖伟[1] 周东辉[1] 孙建风[1] 徐志强[1] 

机构地区：[1]中国海洋大学电子系,山东青岛266071

出　　处：《电子学报》2005年第9期1720-1722,共3页Acta Electronica Sinica

摘　　要：针对移动机器人建立了基于BP神经网络的智能避障控制模型,提出了初始权值优化技术,使得样本组与初始权值相匹配,显著地提高了网络的收敛速度.为了提高系统的实时性,文中采用C和汇编语言混合编制控制程序.计算机仿真和实测结果表明该系统具有学习能力强、人机交互效果好等优点.An intelligent obstacle avoidance model based on BP neural network is established,Also a novel optimal weights initialization technology is proposed so that the sample sets and initial weights can match perfectly. Consequently, the convergence speed increases evidently. In order to improve the real-time performance, hybrid programming using C and assemble language is adopted. Computer simulation and real test show that the system has a strong ability of learning and good performance of human computer interaction.

关 键 词：移动机器人 避障 神经网络 BP算法 初始权值优化技术 

分 类 号：TP24[自动化与计算机技术—检测技术与自动化装置]