Improved Distributed Particle Filter for Simultaneous Localization and Mapping

The Simultaneous localization and mapping (SLAM) problem have become a focus of many researches on robot navigation. Generally the most widely used filter in SLAM problems are centralized filter. It is well known that SLAM based on conventional centralized filter must reconfigure the entire state vectors when the observation dimension changes, which cause an exponential growth in computation quantities and difficulties in isolate potential faults. In this paper, we proposed improved DPF distributed particle filter-SLAM in two aspects, in DPF-SLAM one centralized filter is divided into several distributed filters which reduce the computation quantities efficiently and avoid the necessary to reconfigure the entire state vectors in every step. First, we improved the important function of the local filters in distributed particle filter. By changed a set constant in the important function to an adaptive value, we improved the robustness of the system. Second, we propose an information fusion method that mixed the innovation method and the number of effective particles method, which combined the advantages of these two methods. The result of simulations shows that the algorithms we proposed improved the virtue of the DPF-SLAM system in isolate faults and enabled the system has a better tolerance and robustness.