Distributed Decision-Making for Robot Teams
Decentralized coordination, active search, and mapping for robot teams operating far beyond reliable communications.
This thread studies how each robot in a team can decide for itself — where to search, what to map, and what to share — so the team stays effective in unknown, hazardous, and communication-denied environments.
When robot teams leave the lab, centralized coordination breaks down: links drop, bandwidth collapses, and no single vehicle ever sees the whole picture. Our work pushes the decision-making onto the robots themselves.
The foundation is communication-efficient perception. By modeling local occupancy as Gaussian mixtures and exchanging only a sparse set of informative views, robots share a global picture of the environment under tight bandwidth budgets. Built on this, our distributed mapping approach lets a multi-robot team rapidly explore subterranean environments at high fidelity, validated in hardware experiments in a wild cave in West Virginia.
More recent work makes the coordination fully decentralized. In active search, each robot biases its trajectories toward resolving uncertainty over the locations of an a priori unknown number of targets, exploiting stochasticity for coverage when communication is denied and fusing detections across vehicles when it is available. Field experiments with a team of aerial robots outperform greedy coverage-based planning in communication-denied scenarios while localizing all targets to within a few meters.
