Strategizing for Cyber Security Enhancement Autonomous Intersection Management (AIM) based on Multi Agent Deep Reinforcement Learning (MADRL)

Investigator(s):

• Yongxin Liu, Embry-Riddle Aeronautical University, ORCID # 0000-0003-4527-8623  (PI)

Project Description:

Figure 1 Reinforcement Learning (RL) enabled AIM defense framework against cyber attacks

We propose a Reinforcement Learning (RL) enabled defense framework to dynamically derive appropriate policies to secure the communication and safety of AIM: first, we propose a digital-twin and multi-agent enabled simulated environment that evaluates the systematic intersection throughputs, communication overhead, potential user privacy disclosure and robustness metrics. Second, we propose to develop a RL agent with explainable AI techniques that can dynamically find the optimal combination of secure properties that can maintain the efficient operation of intersections under ongoing or potential cyber-attacks. Third, we propose using RL to find adaptive anonymization techniques for data shared between vehicles and infrastructure. The system could learn the appropriate level of anonymization needed to maintain privacy while still allowing for effective traffic management.