This paper studies online multi-robot routing when some robots spoof their own localisation. It models a monitor-aware adversary that keeps its spoofing within a distance limit, and a monitor that combines localisation trust (calibrated on real GPS spoofing data) with behavioural evidence from task execution, then removes flagged robots from planning. Undetected adversaries break the planner-execution consistency that rollout depends on, so rollout loses its usual cost improvement; removing them restores it, in experiments on real spoofing datasets and San Francisco taxi demand.
Relevance
A rare paper that follows a compromised teammate past the estimator and into the planner. It fits robust robotics and multi-robot systems; the open question is how fast detection must be before the routing cost cannot be recovered.
Links
- arXiv
- Subfield: Security, Planning and navigation, Multi-robot systems