Persistent Surveillance of Events with Unknown Rate Statistics
We present a novel algorithm for persistent monitoring of stochastic events that occur at discrete locations in the environment with unknown event rates. Prior research on persistent monitoring assumes knowledge of event rates, which is often not the case in robotics applications. We consider the multi-objective optimization of maximizing the total number of events observed in a balanced manner subject to real-world autonomous system constraints. We formulate an algorithm that quantifies and leverages uncertainty over events’ statistics to greedily generate adaptive policies that simultaneously consider learning and monitoring objectives. We analyze the favorable properties of our algorithm as a function of monitoring cycles and provide simulation results demonstrating our method’s effectiveness in real-world inspired monitoring applications.
