Creating Visual Reactive Robot Behaviors Using Growing Neural Gas.

Abstract

Creating reactive robot behaviors that rely solely on vi-sual input is tricky due to the well-known problems in-volved with computer vision. This paper presents a po-tential solution to this problem. The robot builds repre-sentations of its target environment using Growing Neu-ral Gas. The robot programmer then specifies its behav-ior for each learned node. This approach is shown to have the potential to be effective for simplifying robot behavior programming in an indoor environment with low-cost hardware. Programming reactive behaviors (Brooks 1986) on a robot equipped with basic sensors such as touch sensors and sonars is reasonably straightforward and well-understood. Despite considerable progress in computer vision tech-