Task-State Integrity Attacks and Defenses for Fixed-Route Autonomous Ice Resurfacing

Abstract

Fixed-route robotic service systems separate navigation from service application. A platform follows a repeated coverage route while adapting a scalar service command from a spatial task-state map, and when that map is corrupted, service quality can degrade even though the route stays safe, deterministic, and collision-free. Task-state integrity is thus a cyber-physical concern distinct from navigation safety. The concern is studied through a reproducible evaluation framework instantiated for autonomous ice resurfacing, in which a resurfacer computes water commands from an observed surface-demand map that an attacker may corrupt. False-positive, false-negative, combined, and kernel-aware map corruptions are evaluated against a naive controller, an uncertainty-aware Bayesian controller, and several low-cost filters. The naive attacks produce measurable excess service, missed service, and readiness delay. Bayesian smoothing mitigates the naive attacks. Kernel-aware false positives yield greater absolute regional excess for Bayesian than naive allocation under budget-matched controls, but smaller attack-induced increases, exposing a distinction between service quality and attack sensitivity. The low-cost filters expose complementary tradeoffs rather than a dominant defense. These results indicate that adaptive robotic service systems must protect task-state integrity in addition to safe coverage and navigation.

Index terms: Autonomous ice resurfacing, cyber-physical systems, false-data injection, task-state integrity

Research context

This paper is the first study in my dissertation on autonomous ice resurfacers. Read more about the project.