Multi-robot simultaneous localization and mapping using particle filters

Abstract — This paper describes an on-line algorithm for multirobot simultaneous localization and mapping (SLAM). We take as our starting point the single-robot Rao-Blackwellized particle filter described in [1] and make three key generalizations. First, we extend the particle filter to handle multi-robot SLAM problems in which the initial pose of the robots is known (such as occurs when all robots start from the same location). Second, we introduce an approximation to solve the more general problem in which the initial pose of robots is not known a priori (such as occurs when the robots start from widely separated locations). In this latter case, we assume that pairs of robots will eventually encounter one another, thereby determining their relative pose. We use this relative pose to initialize the filter, and combine the subsequent observations from both robots into a common map. Third and finally, we introduce a method for integrating observations collected prior to the first robot encounter, using the notion of a virtual robot travelling backwards in time. This novel approach allows us to integrate all data from all robots into a single common map. I