Decentralized Ergodic Coverage Control in Unknown Time-Varying Environments

A team from UC Berkeley, led by Maria G. Mendoza, has published a novel framework for coordinating autonomous drone swarms in chaotic, evolving disaster scenarios. The core challenge they address is maintaining situational awareness in environments where key areas—like a spreading wildfire front or a flood zone—are constantly changing. Traditional methods often fail because they assume a static map or require a central command hub, which can be a single point of failure. This new 'decentralized ergodic coverage control' strategy gives each drone (agent) the intelligence to independently decide where to go next.

Each drone uses a mathematical model called a Gaussian Process to continuously update its 'belief' about the environment's importance map—essentially, its best guess of where critical activity is happening. It then follows an 'ergodic' policy, implemented via a Markov-chain model, which drives it to spend time in areas proportional to their estimated importance. This creates an emergent swarm behavior that efficiently balances deep monitoring of known hotspots with broad exploration to detect new threats. The system was tested in simulations of disaster evolution and showed superior adaptability and transient performance compared to existing coverage strategies, all without any agent needing a complete picture or central instruction.