A neural implicit field BEV framework for road elevation estimation with application to active suspension preview
This paper presents a neural implicit field BEV framework for local road-elevation estimation from two consecutive multi-camera frames. Relative ego pose aligns historical BEV features to the current ego frame before fusion, and a coordinate-conditioned implicit decoder predicts local-plane residual elevation continuously in a forward ROI. LiDAR supervision uses a lower-envelope road-surface procedure and a valid-cell mask. On 6019 nuScenes validation samples (698,699 valid cells), the method achieves 29.02 mm MAE, 65.22 mm RMSE and 23.96 mm wheel-track MAE. A standard fixed-weight MPC and four-degree-of-freedom half-car numerical study over 14 independent profiles shows that BEV preview reduces mean body-acceleration RMS by 1.82% relative to flat-preview MPC and 11.25% relative to passive suspension. The proposed framework is evaluated for road-elevation estimation and its potential to provide preview information in a numerical active-suspension study.
Authors
- Xinxin Jiang (ORCID: https://orcid.org/0000-0003-0791-5812)
- Rui Zhao (ORCID: https://orcid.org/0000-0002-9401-9575)
Institutions
- Nanjing University of Aeronautics and Astronautics (CN)
Publication Details
- Journal
- Proceedings of the Institution of Mechanical Engineers Part D Journal of Automobile Engineering
- Published
- 2026-11-19
- DOI
- https://doi.org/10.1177/09544070261507629
- Primary Topic
- Autonomous Vehicle Technology and Safety
- Type
- article
- Field-Weighted Citation Impact
- 0.00