Abstract
This paper addresses the cooperative localization problem for a multiagent system in the framework of belief propagation. In particular, we consider the RoboCup 3D Soccer Simulation scenario, in which the networked agents are able to obtain simulated measurements of the distance and bearing to both known landmarks and teammates as well as the direction of arrival (DOA) of messages received from allies around the field. There are, however, severe communication restrictions between the agents, which limit the size and periodicity of the information that can be exchanged between them. We factorize the joint probability density function of the state of the robots conditioned on all measurements in the network in order to derive the corresponding factor graph representation of the cooperative localization problem. Then we apply the sum-product-algorithm (SPA) and introduce suitable implementations thereof using hybrid Gaussian-Mixture Model (GMM) / Sequential Monte Carlo (SMC) representations of the individual messages that are passed at each network location. Simulated results show that the cooperative estimates for position and orientation converge faster and present smaller errors when compared to the non-cooperative estimates in situations where agents do not observe landmarks for a long period.
Highlights
In several modern engineering applications, multiple agents with sensing, processing, communication and actuating means cooperate towards a common objective
We focus on the cooperative localization problem considering a sensing environment based on the RoboCup 3D Soccer Simulation domain [16], in which individual humanoid soccer players are able to indirectly observe their corresponding locations with respect to (w.r.t.) known and uniquely identified landmarks, e.g. soccer field corners or goalposts, and w.r.t. allied players1
We introduce the proposed factorization of the joint posterior distribution of the individual robot poses conditioned on all available network measurements and derive its corresponding factor graph (FG) representation
Summary
In several modern engineering applications, multiple agents with sensing, processing, communication and actuating means cooperate towards a common objective. The communication limitation imposed by the rules is a major concern, since the agents only transmit unidirectionally, at pre-allocated time slots and a limited number of bytes per slot An algorithm for such a scenario cannot require multiple information exchange between robots at the time instants where mutual measurements are assimilated. Compared to non-cooperative methods for localization, which rely solely on local data, the incorporation of mutual measurements to the self-localization problem reduces the number of detected landmarks required by a robot to achieve a single modal estimate and, as verified in this paper, leads to a significant improvement in the convergence of the errors and to higher accuracy in scenarios where the robots remain a long time without detecting any landmarks. The p.d.f. of a multivariate real-valued Gaussian distribution parameterized by the mean vector μ and the covariance matrix Σ is represented as fN (x; μ, Σ)
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
