Privacy-Preserved Federated Reinforcement Learning for Autonomy in Signalized Intersections
Negar Asadi, Seyed Hamid Hosseini, Mahdi Imani, Daniel P. Aldrich, Seyede Fatemeh Ghoreishi
Abstract
The control of autonomous vehicles (AV) within complex signalized intersection scenarios poses a substantial challenge, requiring a significant amount of data for effective training of autonomous driving models. This challenge is compounded by limitations in relying solely on a single AV for data, which may fall short of capturing the diverse and dynamic nature of complex intersection settings. The advent of connected and autonomous vehicles (CAVs) introduces a transformative concept, enabling real-time traffic information exchange among interconnected vehicle networks. This shift empowers CAVs to exhibit vastly improved driving behaviors, leveraging shared information for enhanced accuracy, reliability, and efficiency. However, the imperative of preserving user privacy and ensuring data security in the context of AI-enabled CAVs remains a significant challenge. This paper proposes a federated reinforcement learning framework designed to establish a privacy-preserving knowledge-sharing strategy. The proposed framework efficiently aggregates and reuses knowledge learned by diverse CAVs, operating in varying intersection environments. This collective knowledge enhances decision-making precision and adaptability, contributing to safer and more efficient autonomous driving at signalized intersections.