Efficient way of path planning and navigation for autonomous multi robot systems using an intelligent vision system

The paper proposes a vision system for robots in a multi robot system that enables the robots to perceive vision by de-centralizing and re-using the learning of each and every robot in the multi robot system. By de-centralized and re-using the learning it is meant that the robots learn about the environment from their own point of view with their own sensors but share their learning with other robots, thus forming a pool of information about the environment in a distributed way. This distributed perception makes the generation of multi view appearance model of unknown, arbitrary objects easy and effective. The paper focuses on developing such a de-centralized system to track multiple obstacles simultaneously which will help in efficient path planning and navigation. The saved learning by one robot can be used by other robots in the multi-robot system. On every run the proposed system continues learning about the objects from where it left. The proposed vision system is capable of tracking objects efficiently at different scales, lighting conditions, environments and also under occlusions. It is implemented on a mobile robot with the low-power OMAP4 based pandaboard ES as the processing unit and the performances are discussed.