University of Southern California, Los Angeles Centre for Robotics and Embedded Systems

Single-robot Markov localization techniques have proven to be robust and effective in real-world environments, in particular sample-based techniques such as Monté Carlo methods. These methods have experimentally been show to successfully localize in realtime onboard an actual robot in real-world environments. Recent work has extended such methods to the multi-robot localization domain with some success [1]. In the multi-robot case, important information is gained when a robot detects another robot, and with care, this information can be incorporated into the probability density function of both the detector and detectee robot. This can lead to an increase in localization speed and accuracy over the single-robot case. An area that has received some attention in the single-robot case [2] and very little attention in the multi-robot case is active localization. In active localization, the robot(s) may actively choose actions so as to aid in localization. Active localization has the potential to further increase the speed and accuracy of localization. We propose to implement a method of collaborative multi-robot active localization and investigate its properties as compared to collaborative multi-robot passive localization. All features of the localization process will be done in a distributed manner with no centralized control or processing of data. We will use the Player/Stage simulation environment to develop and explore the characteristics of multi-robot active localization as compared to the passive localization case. We will validate the results from the investigations in simulation through the use of real-world experiments using N Pioneer 2DX robots. A key requirement for multi-robot localization is the ability for the robots to detect other robots that are then incorporated into the robots pose belief. To implement this feature, we will equip each robot a laser beacon and an accompanying unique color barcode. The laser beacon will enable the robots to detect the presence of another robot and the unique color barcode will enable the determination of the identification of the detected robot. This has been used in other existing experimental work in the field, for example [3]. Each robot will independently run Markov localization using Monté Carlo sample-based methods to represent its own belief function. We will use standard Monté Carlo localization by augmenting the observation model with the addition of an observation component representing the detection of another robot to implement collaborative multi-robot localization, a technique inspired by [1]. Each robot will only maintain its own belief function as maintaining the joint distribution of the belief functions of all robots in not feasible. When a robot detects another robot, the current population of samples from the detected robot will be given to the detectee and used in conjuction with the observation to update its (the detectee’s) current belief. The symetric case will also be done, the current population of samples from the detectee will be given to the detected robot and used to update its belief function. We are not decided at this time on how we will make use of the active component of the localization process. In general, each robot will do some computation on its current belief function and use the results of this computation to decide upon future actions. The actions may be to take specific actions itself, or it could be a request to another robot(s) to make specific actions to help disambiguate its own pose, or both. Whatever mechanism we use to implement active localization, we aim to keep all computations distributed and executable on the actual robots in a timely manner. It is believed that through communication of robots believed locations, and associated certainties, several techniques can be used to minimise the total uncertainty within the group through collaboration. Thus the multi-robot active case is significantly more complicated than the single agent versions considered in the literature [2].