Secure Planning Against Stealthy Attacks via Model-Free Reinforcement\n Learning

摘要

We consider the problem of security-aware planning in an unknown stochastic\nenvironment, in the presence of attacks on control signals (i.e., actuators) of\nthe robot. We model the attacker as an agent who has the full knowledge of the\ncontroller as well as the employed intrusion-detection system and who wants to\nprevent the controller from performing tasks while staying stealthy. We\nformulate the problem as a stochastic game between the attacker and the\ncontroller and present an approach to express the objective of such an agent\nand the controller as a combined linear temporal logic (LTL) formula. We then\nshow that the planning problem, described formally as the problem of satisfying\nan LTL formula in a stochastic game, can be solved via model-free reinforcement\nlearning when the environment is completely unknown. Finally, we illustrate and\nevaluate our methods on two robotic planning case studies.\n