Multi-Agent SLAM with Consensus Mapping on Graph

Abstract

In this article the problem of tracking the state of the localization system in the underground mine is discussed. This state consists of positions of the access points of the system in the environment of the underground mine. Due to installation and maintenance errors the real state of the localization system may differ from the expected one. Previously published algorithm showed sufficient accuracy but had drawbacks, one of which is requirement to cover big areas in a single run. In this article a new solution is proposed. Previously used FastSLAM-based approach is expanded to track several independent mobile agents each of which builds an independent map of the access points. These maps are periodically combined through the chosen opinion pooling method. Opinion pooling is a family of methods of combining several probabilities of an event measured by different observers to create a single estimation of the probability. Several such methods were evaluated in this article. Proposed multi-agent solution enhances flexibility of the algorithm which allows splitting up the big problem of estimating the state of the system into several smaller ones. Proposed solution was evaluated on a synthetic dataset. The experimental results showed that multi-agent approach raises mean mapping error but is applicable in general.