CONSENSUS PARTICLE FILTERING ON DISTRIBUTED MULTI-ROBOT SLAM

• Main Author: HIDAYAT - NIM: 33211305 , FADHIL
• Format: Dissertations
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/27056
• Institution: Institut Teknologi Bandung

<p align="justify">Extensive research on single-robot simultaneous localization and mapping (SLAM) over the past decade has provided good estimation results in mapping small environment. This raises the idea for building a larger map by using some robots that are assembled within the team, also called multi-robot SLAM. There are two types of multi-robot SLAM, that is, centralized and distributed multi-robot SLAM. Centralized multi-robot SLAM method still has problems in case of system failure on some robots, especially the system failure on the robot central. Distributed multi-robot SLAM was developed to overcome the weaknesses in centralized multirobot SLAM by: 1) building a map of larger environment by fusing local maps from a group of robots, 2) eliminating the dependence on central processing by using distributed computing, 3) overcoming the vulnerability in case of system failure in some robots, especially the system failure in the robot central, and 4) eliminating centralized communication, in which each robot requires only local communication with its neighboring robots. However, the distributed multi-robot SLAM method has not yielded good results on map estimates and localization estimates, and map convergence still has problem at loop closure. <br /> <br /> This research proposes consensus-based distributed multi-robot SLAM method that can be applied to map common environments. This research use FastSLAM algorithm as the main base in developing distributed multi-robot SLAM. This research suggests using one of two selection of consensus parameters, that is, particle weight and posterior parameter. This research assume that each robot has the same motion model and observation model; therefore, the system noise and observation noise is the same if applied to a common environment. The aim of this research is to design consensus based distributed multi-robot SLAM method that provides better map estimation and localization results when applied to common environments. Communication between robot is ad-hoc communication. Each robot broadcasts its local information, another robot captures information broadcasted using ad-hoc communication. Signals strength is used as a priority measure of data exchange between robots. <br /> <br /> This research concentrates to test the proposed methods on aspects: 1) root mean square error (RMSE) of map to see the map estimation performance, 2) RMSE of localization to see the robot’s pose estimation performance, 3) processing time to see how long the computation process takes place per timestep, and 4) map convergence at loop closure. <br /> <br /> The results of this research are the proposed methods deliver better performance in terms of map and localization estimation, and map convergence when compared to distributed multi-robot SLAM method. The performance improvements also resulted in an increase in computation time.<p align="justify">