LOGIC+: LiDAR-Only Geometric-Intensity Confidence Grids for Drivable Area Estimation
Autonomous vehicles today rely on high-definition maps for navigation and scene understanding. The creation and maintenance of these maps are costly processes that raise the entry bar for the deployment of autonomous driving technologies in the real world. A potential solution to this problem is est...
| Autores: | , , |
|---|---|
| Formato: | artículo |
| Estado: | Versión publicada |
| Fecha de publicación: | 2026 |
| País: | España |
| Recursos: | Consejo Superior de Investigaciones Científicas (CSIC) |
| Repositorio: | DIGITAL.CSIC. Repositorio Institucional del CSIC |
| OAI Identifier: | oai:digital.csic.es:10261/423803 |
| Acesso em linha: | http://hdl.handle.net/10261/423803 |
| Access Level: | acceso abierto |
| Palavra-chave: | Autonomous vehicles perception LiDAR drivable area road detection dynamic occupancy grid |
| Resumo: | Autonomous vehicles today rely on high-definition maps for navigation and scene understanding. The creation and maintenance of these maps are costly processes that raise the entry bar for the deployment of autonomous driving technologies in the real world. A potential solution to this problem is estimating the drivable area in real time, a capability made possible by recent advancements in sensor technology and particularly relevant for complex urban environments. LiDAR-only methods for detecting drivable area are scarce and typically appear in fusion frameworks with other sensor technologies. Nevertheless, the optimization of single-sensor modalities coupled with flexible fusion solutions are key to unlock the dependencies on high-definition maps that navigation systems have nowadays. In this work we propose LOGICC: a LiDAR-Only Geometric-Intensity Confidence Grids drivable area estimation algorithm. The approach leverages both local and non-local geometric features of point clouds, using non-parametric techniques for intensity analysis. These features are treated as individual drivability estimations and computed with confidence maps that allow for intelligent fusion in a Linear-Opinion Pool. The fused drivability proposals are combined with occupancy information and input into a Dynamic Occupancy Grid to handle moving obstacles in the environment. The proposed method is tested in the Waymo Open Dataset which includes diverse urban driving scenes where is able to match the performance of state-of-the-art approaches without training or case-by-case parameter tuning. |
|---|