Adaptive Macroscopic Ensemble Allocation for Robot Teams Monitoring Spatiotemporal Processes

Advances in real‐time data processing have enabled robot teams to continuously adapt their sampling locations as they monitor their environments, enabling them to build highly predictive models of complex, dynamic environments. These models, in turn, enable robots to make better plans and more quickly adapt to changing environmental conditions. However, continuously identifying high information content regions and assigning robots to these new locations requires addressing the long‐standing multi‐robot task allocation problem. Existing allocation methods use task‐specific planning and/or control strategies that lack the flexibility needed to monitor spatiotemporal environments. In contrast, biological collectives robustly handle a wide range of environment conditions by relying on resource selection mechanisms that are beneficial to the survival of the population. Taking inspiration from biology, we address the challenge of flexibly monitoring spatiotemporal environments by using a team‐wide macroscopic ensemble approach which naturally mimics biological selection techniques. Existing macroscopic allocation strategies enable robots to switch between sampling regions, but unlike biological counterparts cannot respond to changing environmental conditions and perform poorly when team sizes are small. In this work, we introduce an online adaptive macroscopic allocation strategy that leverages environmental feedback to enable adaptation to changing environmental conditions. Our approach results in the synthesis of single‐agent task selection policies that achieves flexible assignment of robots for a range of dynamic conditions that perform well even when team sizes are small.