Trust-Evaluation-Based Intrusion Detection and Reinforcement Learning in Autonomous Driving

With ever-increasing penetration of autonomous driving vehicles (ADVs), security becomes of significant essence to autonomous vehicular networks (AVNs). On one hand, although cryptographic systems help deal with various attacks, they cannot resolve inside attacks from compromised or malfunctioning ADVs. On the other hand, ADVs with higher automated levels are easier to attack because the automated components onboard are vulnerable to inside attacks. These issues can be mitigated well by trust-based intrusion detection. Therefore,this article presents an efficient trust-based intrusion detection framework for AVNs. First, we propose a novel trust evaluation model for ADVs, in which all the trust evaluation information of a given ADV is utilized to compute its trust value. Then, based on the trust evaluation model, a two-level intrusion detection framework is presented. In the framework, the trustworthiness of an accident or attack warning is established not only based on trust evaluation with the coverage of a roadside unit (RSU), but also the information exchanged between RSUs through the cloud server. Afterward, we propose a reinforcement-learning-based incentive mechanism to stimulate ADVs to report warnings. Through the case study, the proposed framework outperforms the conventional mechanism and can reach a higher warning detection ratio than the conventional methods.