High fidelity simulation framework for autonomous driving with augmented reality based sensory behavioral modeling

This study addresses the challenge of simulation fidelity in autonomous driving by proposing a high -fidelity simulation framework enhanced with augmented reality. The framework integrates Generative Adversarial Networks (GANs) and Neural Radiance Fields (NeRF) for high-fidelity sensory data synthesis, employs reinforcement learning for dynamic behavior modeling, and implements Hardware-in-the-Loop (HIL) testing for real-time interaction. Experimental results demonstrate significant achievements in visual fidelity, with pixel intensity averaging 120.3±15.2 in clear weather conditions, while tex ture gradients (1.8±0.3) and edge sharpness (0.85±0.05) remain consistent across various scenarios; in dynamic behavior modeling, the system achieves response times of 0.85±0.14 seconds, following distances of 5.2±0.7 meters, and minimal lane deviation of 0.22±0.03 meters. The findings indicate that this framework effectively bridges the simulation-to-reality gap, providing a reliable simulation environment for autonomous driving algorithm training and validation.