Resilient Decentralized Ergodic Coverage for Scalable Multi-Robot Systems in Unknown Time-Varying Environments
Maintaining situational awareness in high-stakes multi-robot applications requires balancing exploration of unobserved regions with sustained monitoring of important regions, often in unknown, time-varying environments, under partial observability and degraded operating conditions. We propose a decentralized, graph-based ergodic coverage framework for persistent, long-horizon monitoring as a high-level planner. The environment is represented as an \textit{importance map}: a distribution over a graph of regions whose weights define how important each region is, with regions of interest (ROIs) denoting those of high importance. Each agent maintains an online Gaussian Process (GP) belief over this map from its own noisy observations and those of its neighbors, and follows an ergodic Markov-chain policy that steers its long-run visitation toward this belief and is updated locally as the belief changes. Unlike existing approaches that assume known importance maps or dynamics, centralized coordination, or a static environment, our framework addresses these challenges jointly. We evaluate the framework in long-horizon UAV disaster-response simulations in which ROIs spread, appear, and relocate, and analyze its computational complexity and belief accuracy. We show that the team keeps covering the map while consistently finding new and shifting ROIs, remains resilient to communication and memory degradation and robot loss, and scales to hundreds of robots.