Decentralized Planning For Active Information Gathering On Targets With Probabilistic Model

Motion strategies for multiple robots actively acquiring information in a dynamic environment have been widely studied. However, existing active information gather- ing algorithms are restricted by assumption of linear target process, or deals with only limited number of agents and targets because of high cost of reward computation. In this paper, we formulated the active information gathering problem with the belief distribution of desired process, instead of a specific model. The reward function is derived based on mutual information of measurement and belief distribution, which can be efficiently computed under the Gaussian assumption on belief distribution. A decentralized path planner is designed to maximize reward function, which scales well in both numbers of agents and targets. We apply the proposed planner to a cooperative localization scenario and validate the performance and scalability in numerical simulation.

Bibtex

–>