Simulating Realistic LiDAR Data Under Adverse Weather for Autonomous Vehicles: A Physics-Informed Learning Approach

2026-04-01

AI Summary

Existing LiDAR simulation methods struggle to accurately model the complex interactions between laser beams and atmospheric particles under adverse weather conditions, leading to significant data distortion. To address this limitation, this work proposes a Physics-Informed Conditional Wasserstein Generative Adversarial Network (PICWGAN), which uniquely integrates a physics-based signal attenuation model and a geometrically consistent degradation mechanism into the generative adversarial framework to enhance the realism of synthetic LiDAR data in rain and snow. The proposed approach substantially reduces the domain gap between simulated and real-world data, producing intensity distributions on the CADC, Boreas, and VoxelScape datasets that closely align with those of actual measurements. Furthermore, 3D object detection models trained with PICWGAN-augmented data achieve performance comparable to models trained exclusively on real data.

Technology Category

Computer Vision: Generative Adversarial Networks (GANs) for VisionNatural Language Processing: GenerationSearch and Optimization: Sampling/Simulation-based Search

Application Category

Web Mining and Content Analysis: Web data generation and simulationSearch and Retrieval-Augmented AI: Large language models for searchSemantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Accurate LiDAR simulation is crucial for autonomous driving, especially under adverse weather conditions. Existing methods struggle to capture the complex interactions between LiDAR signals and atmospheric phenomena, leading to unrealistic representations. This paper presents a physics-informed learning framework (PICWGAN) for generating realistic LiDAR data under adverse weather conditions. By integrating physicsdriven constraints for modeling signal attenuation and geometryconsistent degradations into a physics-informed learning pipeline, the proposed method reduces the sim-to-real gap. Evaluations on real-world datasets (CADC for snow, Boreas for rain) and the VoxelScape dataset show that our approach closely mimics realworld intensity patterns. Quantitative metrics, including MSE, SSIM, KL divergence, and Wasserstein distance, demonstrate statistically consistent intensity distributions. Additionally, models trained on data enhanced by our framework outperform baselines in downstream 3D object detection, achieving performance comparable to models trained on real-world data. These results highlight the effectiveness of the proposed approach in improving the realism of LiDAR data and enabling robust perception under adverse weather conditions.

Problem

• LiDAR simulation
• adverse weather
• sim-to-real gap
• realistic data generation
• autonomous vehicles

Innovation

• physics-informed learning
• LiDAR simulation
• adverse weather
• sim-to-real gap
• 3D object detection