Vehicle-to-Infrastructure Multi-Sensor Fusion (V2I-MSF) With Reinforcement Learning Framework for Enhancing Autonomous Vehicle Perception

Traditional autonomous driving systems, which rely solely on vehicle onboard unit (OBU) sensors, often face challenges such as limited perception range and sensor occlusion, while systems dependent only on roadside infrastructure sensors may lack the fine-grained data required for accurate, vehicle-level decision-making. To address these challenges, this paper presents a Vehicle-to-Infrastructure Multi-Sensor Fusion (V2I-MSF) framework that integrates LiDAR and camera data from vehicle OBUs with sensor data from roadside infrastructure. This fusion enhances environmental visualization and supports more informed decision-making for autonomous vehicles. In addition, we propose an Actor-Critic reinforcement learning (RL) model designed to process the fused sensor data, enabling precise motion estimation, obstacle avoidance, and lane prediction. The proposed framework demonstrates significant improvements in accuracy, extended perception range, and robustness in decision-making, particularly in complex urban intersection scenarios. Simulation results validate the effectiveness of the V2I-MSF approach, showing its superiority over standalone sensor configurations in creating safer, more efficient driving environments.