Kherroubi, Zine el abidine (Technology Innovation Institute)

A Generalizable Actor-Critic Framework for Safe and Robust Driving in V2X Environments

Scheduled for presentation during the Regular Session "S40a-Cooperative and Connected Autonomous Systems" (FR-LM-T40), Friday, November 21, 2025, 10:50−11:10, Cooleangata 4

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on October 18, 2025

Abstract

Connected and autonomous vehicles (CAVs) leverage V2X communication to enhance situational awareness and cooperative driving. However, V2X networks are prone to impairments such as data delays and losses, leading to impaired observability that degrades control performance. Addressing this challenge, we present the Blind Actor-Critic, a reinforcement learning algorithm designed to ensure robust decision making under V2X impairments. Building upon a unified actor-critic architecture, this study focuses on demonstrating the generalizability of the Blind Actor-Critic mechanism across a range of actor-critic frameworks, comprising TD3, SAC, TRPO, and PPO. Through extensive simulations in driving scenarios under varying V2X reliability conditions, we illustrate that integrating the Blind Actor-Critic mechanism consistently improves training stability by reducing the residual variance of the value function and enhancing reward learning. Moreover, it maintains robust control across all evaluated frameworks when tested across challenging driving scenarios with unreliable V2X communication. These results confirm the versatility and effectiveness of the Blind Actor-Critic as a generalizable solution for reinforcement learning in impaired V2X environments.