Abstract
In recent years, demands for advanced sensing in robot tasks using multiple sensing modules have been increased. In robot vision tasks, distributed use of multiple cameras is required to overcome limitations of a narrow field-of-view and low 3D resolution when using a single camera. If a target object moves outside the field-of-view of a camera, we need to automatically switch the tracking task from the camera to another module. Integration of multiple sensing information is also an important issue. A traditional approach for construction of such system is to organize the modules with a single and centrally controlled system. However, this approach has disadvantages in various aspects: flexibility, robustness, extendability, and efficiency in achieving tasks. To overcome these difficulties, decentralized and cooperative control of multi-robot is a promising approach. In this paper, we propose a concept of decentralized and cooperative sensing system for robot vision tasks. We present a prototype of such visual sensing system implemented using a real multi-robot system. It performs visual inspection of water leakage on a sleeve of a valve. A novel feature of the system is it consists of 3D model-based agents which communicate 3D model information of the task environment for distributed sensor planning.
