Ultra-High-speed Sensory-Motor System Based on Distributed Network Architecture

本研究的目的在于开发一种能在真实环境中灵活高速行动的识别-行为整合系统。利用基于分层并行架构的感觉运动集成、超高速识别与运动、分布式网络处理架构等基础技术，我们开发了一种基于分布式网络架构的大量传感器和执行器连接的高速实时处理系统。具有TCP/IP通信功能的高速视觉网络系统。为了在无遮挡的情况下对大面积高速运动的多目标进行跟踪，我们开发了一种多视觉系统的分布式协同网络系统。一种分布式传感器网络信息处理算法。我们提出了一种传感器融合处理方法DTKF （delay - tolerance Kalman Filter）。DTKF是一种利用卡尔曼滤波的方法，随着传感器数量的增加，该方法具有很好的可扩展性。通过高速视觉传感器目标跟踪任务的数值仿真验证了该方法的有效性。一种多高速视觉反馈系统。我们开发了一种多高速视觉反馈系统。通过插值多个视觉系统的信息缺失，实现了对多目标的三维跟踪，实现了对遮挡的鲁棒性。同时，提高了测量范围和精度，避免了遮挡。基于分层并行处理架构的感觉-运动融合为了连接大量的传感器系统和机器人系统，我们开发了一种将实时处理系统、机器人系统和传感器系统集成在一起的并行分布式处理系统。在该系统中，三个由dSpace公司生产的实时处理系统通过CAN总线相互连接，两个机械臂，三个机械手，三个高速视觉系统（六个摄像头）和其他传感器（触觉传感器，力传感器等）通过高速IO端口连接。少

The purpose of this study is to develop a recognition-behavior integration system which can act in a flexible manner at high-speed in the real environment. Using basic technologies, such as sensory-motor integration based on hierarchical parallel architecture, ultra-high-speed recognition and motion, and distributed network processing architecture, we have developed a high-speed realtime processing system in which a large number of sensors and actuators are connected based on distributed network architecture.1.A high-speed vision network system with TCP/IP communication function.In order to track many targets which moves at high-speed in a wide area without occlusion, we have developed a distributed cooperative network system with multiple vision systems.2.A distributed sensor network information processing algorithm.We have proposed a sensor fusion processing method called DTKF (Delay-Tolerant Kalman Filter). DTKF is a method using Kalman Filter, and it has a good scalability against … More the increase of the number of sensors. The proposed method was verified by a numerical simulation of a target tracking task of high-speed vision sensor.3.A multi-high-speed visual feedback system.We have developed a multi-high-speed visual feedback system with multiple high-speed visions. By interpolating the lack of information using several vision systems, a 3D tracking of multiple targets which is robust against occlusion is achieved. Also, the improvement of a range of measurement and accuracy, and occlusion avoidance is achieved.4.Sensory-motor fusion based on hierarchical parallel processing architectureIn order to connect a large number of sensor systems and robot systems, we have developed a parallel distributed processing system in which realtime processing systems, robot systems, and sensor systems are integrated. In the system three realtime processing systems made by dSpace Inc, are connected to each other using the CAN bus, and two robot arms, three robot hands, three high-speed vision systems (six camera heads), and other sensors (tactile sensors, force sensors, and so on) are connected with high-speed IO ports. Less