Distributed panoramic sensing in multiagent robotics

Abstract

In the field of multiagent robotics, much emphasis has been placed on explicit communication strategies between robots, and little on sensing. Sensing is often used on individual robots for avoiding obstacles, however the authors believe its use at a higher level can be advantageous for multiagent cooperation. The authors introduce a concept of cooperation by observation that features the use of panoramic vision sensing applied to multiagent robotics. Using panoramic vision techniques, a robot can view 360/spl deg/ around itself and also obtain coarse range information to objects in its environment. By also observing other robots in the robot society, localizing oneself within the group can be achieved. Thus each robot can observe its local area and create a local map. These distributed local maps centered around each robot can then be integrated into a larger global map based on the relative localization information between robots. Further, by sensing the positions of other robots and objects, a set of simple behaviors can be used to effectively explore the environment. Each robot calculates the range uncertainty to objects while viewing, and then moves to minimize that uncertainty. Multiagent exploration based on these behaviors has been shown to be effective through preliminary experimentation.<>