Bandit-based multi-agent search under noisy observations


Autonomous search using teams of multiple agents need tractable coordination strategies between the search agents. The strategy must lower the time to identify interesting areas in the search environment, lower the costs/energy usage by the search agents during movement and sensing, and be resilient to the noise present in the sensed data due to the use of low-cost and low-weight sensors. We propose a data-driven, multi-agent search algorithm to achieve these goals using the framework of thresholding multi-armed bandits. For our algorithm, we also provide finite upper bounds on the time taken to complete the search, on the time taken to label all interesting cells, and on the economic costs incurred during the search.


  • Related News & Events