Abstract:

We consider the problem of data-driven spatiotemporal monitoring using a team of mobile robots. Specifically, we model the phenomena being monitored as unknown functions that may vary in space and time. Given limited prior information, we require an energy-constrained team to collect data and identify all critical events, i.e., locations and time instants at which the unknown function crosses a pre-specified threshold. We consider a team that consists of mobile sensors (e.g., drones) and charging stations (e.g., ground carrier vehicles). We balance the monitoring requirements with energy and dynamics constraints on the team using an anytime, optimization-based, iterative approach that updates the robot deployments based on data collected online. We also provide sufficient conditions under which our approach yields a non-trivial collection of space-time intervals that include all critical events. We demonstrate our approach using hardware experiments with drones and provide an empirical analysis of its performance and scalability in simulations.