Dynamic Games of Asymmetric Information for Deceptive Autonomous Vehicles.

摘要

This paper studies rational and persistent deception among intelligent robots to enhance the security and operation efficiency of autonomous vehicles. We present an N-person K-stage nonzero-sum game with an asymmetric information structure where each robot's private information is modeled as a random variable or its type. The deception is persistent as each robot's private type remains unknown to other robots for all stages. The deception is rational as robots aim to achieve their deception goals at minimum cost. Each robot forms a belief on others' types based on state observations and updates it via the Bayesian rule. The level-t perfect Bayesian Nash equilibrium is a natural solution concept of the dynamic game. It demonstrates the sequential rationality of the agents, maintains the belief consistency with the observations and strategies, and provides a reliable prediction of the outcome of the deception game. In particular, in the linear-quadratic setting, we derive a set of extended Riccati equations, obtain the explicit form of the affine state-feedback control, and develop an online computational algorithm. We define the concepts of deceivability and the price of deception to evaluate the strategy design and assess the deception outcome. We investigate a case study of deceptive pursuit-evasion games and use numerical experiments to corroborate the results.