Qualitative Planning in Imperfect Information Games with Active Sensing and Reactive Sensor Attacks: Cost of Unawareness

Synthesizing Attack-Aware Control and Active Sensing Strategies under Reactive Sensor Attacks

“…Given a set Γ of sensors, the set of observations O can be constructed for each state s ∈ S, sensing action σ ∈ Σ and β ∈ B (see Alg. 1 in [6]).…”

“…We consider the probabilistic planning of an autonomous agent (called Player 1, or P1, pronoun 'she') in an adversarial environment, where an attacker (called player 2, or P2, pronoun 'he') can carry out attacks on the sensors of P1. The interaction between P1 and P2 is captured using the following model, introduced in our prior work [6]. • F ⊆ S is a set of final states describing P1's reachability objective.…”

“…3) P1 knows that the sensors are being attacked by P2. The difference between this work and [6] is that the latter does not assume 3) and consider attack-unaware P1. We are interested in solving the following problem: Problem 1.…”

“…The defender receives observation from the uncompromised sensors and has a reachability objective which can be satisfied by reaching a set of goal states. In our previous work [6], we analyze the damage of covert attacks and the cost of attack-unaware control where the defender mistakes the compromised sensors as probabilistic sensor failures. This work investigates the synthesis of attack-aware controller with active perception and control.…”