Towards Multi-robot Formations: Study On Vision-based Localization System

Over the past decade,cooperation of swarm of mobile robots has been widely studied for high error tolerance,efficiency,and scalar extendibility.A multi-robot system has exhibited advantages for various services and surveillances in an unknown environment.Examples include cooperative manipulation,building measurement,and cooperative mapping.In many of these applications,the robots in a team are controlled as a whole to follow required formations to accomplish tasks with satisfied overall performance.Localization is a key technology to address how the robots localize themselves in the operating environment and how they obtain their individual poses with respect to the team.Most of the present approaches put the focus on the local localization only, and few of them discuss how to globally localize the robots in the multi-robot formations.For simplicity,many works just assume that the robots employ absolute positioning capability.This thesis study aims to develop a new framework for global localization of a multi-robot formation system in an unknown indoor environment. Both global and local localizations are addressed in this research.Two main works reported in this thesis are localization system and localization strategies.First,an efficient ceiling vision based localization framework is developed for multi-robot formations.A camera pointing upward to ceiling is installed on each robot such that there is no need to use expensive and hardly calibrated sensing equipment (i.e.,laser sensors and omni-directional cameras).The majority of existing works using the ceiling vision have concentrated on the single robot applications only.This thesis makes an extension of applying the ceiling vision system to multi-robot formations,using newly developed feature detection and data association technologies. Match-based local localization is used to calculate the relative poses amongst the robots,and simultaneous localization and mapping(SLAM) is utilized for global localization.Data association,a key issue of SLAM process,is well considered in this study.For balance of robustness and implementation simplicity,a modified JCBB (Joint Compatibility Branch and Bound) is developed to be used to obtain an optimistic feature match hypothesis quickly and accurately.Second,three localization strategies are proposed for global localization of a multi-robot formation system,to meet different task and environment requirements. The first strategy is to globally localize one robot only(i.e.,leader) and then localize the others based on relative poses amongst the robots.This strategy is relatively easy to implement since only one robot needs to be localized globally.The robots must work physically close enough such that they have overlapped ceiling observations for local localization.The second strategy is that each robot localizes itself individually by implementing SLAM.Since all the robots are facilitated to process SLAM independently,they are not required to work close physically.With this strategy,each robot has its own localization frame,and it is necessary to unify localization results of all the robots in the common world frame.The third strategy is to utilize a common SLAM server,which may be installed on one robot,to globally localize all the robots simultaneously,based on a shared global map.This strategy does not require every robot to have SALM processing capability,but has relative high demand for the common SLAM server to accomplish the complex task.Selection of strategy will be based on actual environmental situation and task requirement.Experiments are finally performed on a group of mobile robots to show that all the three strategies exhibit good localization performance.This research provides a vision-based solution for the localization of indoor multi-robot formations.The proposed visual localization methodologies have potential prospects not only in multi-robot formations,but also in other general multi-agent tasks.The research outputs will eventually benefit the automatic applications in indoor environment,such as hospital,airport,depository and shopping mall,and release people from laborious works.