Applying Panoramic Sensing to Multi Agent Robotic Systems

该研究方法关注的是移动机器人的全向视觉有效性，移动机器人构成了一个多机器人系统。利用全向视觉，机器人可以识别其他机器人并估计它们的相对位置。针对使用全向视觉的多机器人系统，我们处理了以下五个问题（全景视觉包括通过摄像机旋转获得的全向视觉信息和通过摄像机线性运动获得的视觉信息）。(1)机器人识别和相对位置估计算法。(2)多机器人执行全局任务的规划方法。(3)低成本、紧凑型全向视觉传感器的发展。(4)将算法和全向视觉传感器应用于实际系统。(5)全方位视觉信息优点的考虑。在执行给定任务之前，机器人需要在全方向上识别出相应的机器人，并估计出它们的相对位置。我们开发的一种算法通过使用一个简单的约束来解决这个问题，即如果三个机器人相互观察，则三个机器人拍摄的图像中两个机器人投影之间的角间隔之和为180度。该算法具有较强的鲁棒性，适用于存在运动和静态障碍物的复杂环境。然后，机器人执行给定的任务。我们开发了一种基于行为的方法作为地图构建任务的规划方法。每个机器人都有简单的行为来探索未知区域，重建局部环境结构并与其他机器人进行通信。由于系统不需要集中控制，因此可以鲁棒地执行任务。最后，我们开发了低成本、紧凑的全向视觉传感器，并将上述算法应用到真实的机器人系统中，验证了该算法在真实系统中的可靠性。我们认为全向视觉传感器及其开发的算法将成为多机器人系统的关键技术。少

This research approach focuses upon effectiveness of omnidirectional vision for mobile robots which constitute a multiple robotic system. With the omnidirectional vision, the robots can identify other robots and estimate their relative positions. We have dealt with the following five issues with respect to the multiple robotic system using the omnidirectional vision (Panoramic vision consists of both omnidirectional visual information obtained by rotating the camera and visual information obtained with linear motions of the camera).(1) Algorithms to identify robots and estimate their relative positions.(2) Planning methods for executing global tasks with multiple robots.(3) Development of low-cost and compact omnidirectional vision sensors.(4) Applying the algorithms and the omnidirectional vision sensor to real systems.(5) Consideration on merits of the omnidirectional visual information.Before executing given tasks, the robots need to identify corresponding robots in the omnidirectio … More nal images and estimate their relative positions. An algorithm which we have developed solves this problem by using a simple constraint that the sum of angular intervals between two other robot projections in images taken by three robots will be 180 degrees if the three robots observe each other. The algorithm is robust and it can be used in complex environments in which moving and static obstacles exist.Then, the robot execute given tasks. We have developed a behavior-based method as the planning method for a map-building task. Each robot has simple behaviors to explore unknown areas, reconstruct local environment structure and communicate with other robots. Since the system does not need centralized controls, it can execute the task robustly.Finally, we have developed the low-cost and compact omnidirectional vision sensors, implemented the above-mentioned algorithms into a real robotic system, and confirmed the reliability for real systems. We consider the omnidirectional vision sensor and developed algorithms will be key technologies for the multiple robotic systems. Less