Robot Teams: Sharing Visual Memories

Abstract

This paper presents, the Growing Neural Gas (GNG), an unsupervised learning algorithm, which allows a team of robots to memorize scenes and collectively create a general understanding of the environment that is easily understood and referenced by humans is presented. Each robot will have its own memory represented by a graph with nodes encoding the visual information of a video stream as a limited set of representative images. GNG are self-organizing neural networks that can dynamically adapt their reference vectors and topology. Frames are sequentially processed by the GNG, automatically generating nodes, establishing connections between them and creating clusters dynamically.We mainly focus on creating a robot team learning mechanism to achieve a distributed system of robots automatically sharing acquired knowledge with others available within the area. This is done using keyframes representing clusters within the robot memory.