Cooperative Global Localization in Multi-robot System

With the development of robotic technology, the application areas of robots have been greatly widened. The robots may serve as assistants (Elena et al., 2004), rescuers (Robert & Arvin, 2003) and explorers (Schenker et al., 2003). In many cases a multi-robot system has to be used, since many tasks cannot be completed with a single robot and multi-robot system can finish the tasks much more efficiently. As in a single robot system, knowing their relative positions and their global positions in the environment are the preconditions for performing tasks and coordination. Since a robot can determine the location of another robot relative to its own when they see each other, both robots can refine their internal beliefs based on the other robot’s estimate. In this way cooperative localization of multiple robots can greatly improve the localization precision and efficiency (Fox et al., 2000; Trawny et al., 2009). An area that has received some attentions in the single-robot case and very little attention in the multi-robot case is active localization (Fox et al., 1998). In active localization, the robot(s) may actively choose actions so as to aid in localization. Active localization has the potential to increase the speed and accuracy of localization further. In this chapter a mechanism of making robots coordinate their action actively during localization is proposed. In order to determine the exploration strategy, the ability to stably track multi-hypotheses of the robot’s own position and his partners’ positions is very important in active localization. However using traditional particle filters for localization tends to produce premature convergence, i.e. the hypothesis represented by particle filters converge to a small area of the state space with high likelihood too quickly. To overcome this problem a new version of particle filters termed co-evolution particle filters (CEPF) is proposed. Another problem of using particle filters in multi-robot localization system is the communication problem. When several robots want to share their estimates they have to transmit a large set of samples from one robot to another, so a great bandwidth is needed. To solve this problem the reduced set density estimator (RSDE) (Girolami & Chao, 2003) is used to estimate the density over robots’ pose, so that only a small sub-set of the original samples should be transmitted between robots, which can reduce the communication data considerably.