HPLaw: Heterogeneous Parallel LiDARs for Adverse Weather in V2V
Yuhang Liu, Xinyue Ma, Xingxia Wang, Boyi Sun, Yutong Wang, Fenghua Zhu, Fei-Yue Wang
Abstract
Parallel LiDAR emerges as an innovative framework for next-generation intelligent LiDAR systems in autonomous driving. In parallel LiDAR research, V2V (Vehicle-to-Vehicle) cooperative perception is a promising technology which can effectively enhance perception range and accuracy through inter-agent information exchange. Currently, sensor heterogeneity remains a critical challenge in V2V. Although some work has made initial attempts to address this issue, existing studies are primarily conducted under ideal clear-weather conditions, ignoring the impact of variable weather factors in real-world applications. In fact, adverse weather has been shown to significantly degrade the performance of LiDAR systems, with the risk of cumulative degradation in V2V. To address this challenge, we first introduce OPV2V-W and V2V4Real-W as new benchmarks to study sensor heterogeneity in V2V under adverse weather. Then we propose the HPLaw architecture (Heterogeneous Parallel LiDARs for Adverse Weather), a self-knowledge distillation method designed to enhance model robustness across varying weather scenarios. HPLaw employs an efficient PF network to facilitate heterogeneous feature fusion and incorporates an SAKD module to extract weather-invariant features. Extensive experiments demonstrate that the student model in HPLaw achieves outstanding performance under all weather conditions, exhibiting remarkable robustness.
BibTeX
@inproceedings{iros2025_hplawheterogeneo,
title = {HPLaw: Heterogeneous Parallel LiDARs for Adverse Weather in V2V},
author = {Yuhang Liu and Xinyue Ma and Xingxia Wang and Boyi Sun and Yutong Wang and Fenghua Zhu and Fei-Yue Wang},
booktitle = {IROS 2025},
year = {2025}
}