The utilization of multi-robot systems has a major advantage when comparing to single robot systems, for example, with multiple robots working together, it has the potential to accomplish a task faster than a single robot. However, when a team of robots is sharing the same worksite, the Simultaneously Localization and Mapping (SLAM) problem becomes much more difficult to resolve because a huge amount of information is needed to be processed as well as analyzed. But on the other hand, multi-robot SLAM can be more efficient if robots can exchange and share information regarding their sensed data properly. In the SLAM problem, especially for Autonomous Underwater Vehicle (AUV) and Unmanned Aerial Vehicle (UAV), it is necessary to include non-linear and non-Gaussian parameters, for which the traditional Kalman Filter (KF) cannot yield ideal solution. In applications involving non-linear and non-Gaussian parameters, Particle Filters (PF), which are based on the concept of Monte Carlo simulation, are more suitable estimation techniques. However, in problems involving multiple dimensions, such as the multi-robot SLAM problem, when a huge number of particles are being used, two problems, namely particle impoverishment and sample size dependency, will occur during the particle updating stage and these problems will become more severe. The problems will reduce the accuracy of the estimation results and resampling algorithms, such as Sequential Importance Sampling, Stratified Resampling and Systematic Resampling are used to alleviate these two problems.
In my master thesis, a novel PF algorithm for tackling the particle impoverishment and sample size dependency problems is being studied and its application in a multi-robot system is examined. In this algorithm, Ant Colony Optimization (ACO) is incorporated into the generic particle filter in order to drive the proposal distribution to approximate the optimal solution. Mathematical proof and results obtained from a single variable estimation problem as well as from the robot localization problem show that, after the ACO optimization, better proposal distribution and more accurate estimation results can be obtained. In order to evaluate the performance of the ACO improved PF (PFACO) when applied to non-linear and non-Gaussian problems, such as the localization and SLAM problem, studies were conducted and utilization of PFACO algorithm for multi-robot systems was introduced. In a multi-robot environment, when two robots encounter, the same information on the same estimation problem represented by the two sets of particles will be re-evaluated based on information conveyed by particles from different sets. The particles are then merged into a single set and in such cases, parallel computing can be applied in order to reduce the processing time. By software simulation, our results are better than those from traditional approaches both in estimation error and execution time.