基于任务的神经网络多传感器数据融合新方法  

A New Task-based Neural Network Method for Multi-sensor Data Fusion

作　　者：游松[1] 程卫星[1] 王田苗[1] 丑武胜[1] 张启先[1] 

机构地区：[1]北京航空航天大学机器人研究所,北京100083

出　　处：《高技术通讯》2001年第7期71-75,96,共6页Chinese High Technology Letters

基　　金：国家自然科学基金资助项目 (69785 0 0 1)

摘　　要：面向遥操作机器人系统对传感和控制的具体要求及条件 ,提出一种基于任务的神经网络多传感器数据融合新方法。根据遥操作过程中的不同任务 ,利用优化的神经网络算法 ,对双视觉、六维力 /力矩、接近觉 (数字 /模拟 )、指端力、关节角度等多种传感器信息进行融合决策 ,并实时地将融合结果反馈回上层控制系统。A new task based neural network method for multi sensor data fusion in the teleoperated robotic system was proposed. Aiming at different teleoperated tasks, an optimized multi sensor fusion algorithm based on neural network was applied to fuse multi sensor information and real timely feedback useful results to high level, including two visions, 6 D force/torque, distance, finger tip force and angle information etc. This article demonstrates our theory and scheme by means of plug in hole experiments and simulation training results. The method has been applied in teleoperated robotic education system based on Internet for a long time and won a good effect.

关 键 词：机器人 数据融合 神经网络 遥操作 传感器 

分 类 号：TP274[自动化与计算机技术—检测技术与自动化装置]