TerrFlat: Physics-Driven Geometry Representation for Structure-Aware Freespace Detection
Jingwei Yang, Liuyi Wang, Mengjiao Shen, Jiayuan Du, Chengju Liu, Qijun Chen
AI summary
Problem
Current freespace detection methods rely heavily on data-driven models that ignore the explicit physical structure of drivable surfaces, causing poor generalization on complex or uneven terrains.
Approach
The authors introduce TerrFlat-Seg, which constructs a physics-driven geometric map encoding lateral smoothness, longitudinal consistency, and vertical deviation, then fuses it with RGB features via a symmetric recalibration module.
Key results
- Proposes TerrFlat, a physics-driven 3D representation encoding lateral smoothness, longitudinal consistency, and vertical deviation
- Introduces SFFM for bidirectional channel-spatial recalibration between geometric and visual features
- Demonstrates consistent performance gains over baselines on KITTI-Road, Semantic-KITTI, and ORFD datasets
- Validates real-world robustness through deployment on an automated guided vehicle platform
Why it matters
Enables autonomous vehicles and embodied agents to navigate complex, non-planar terrains more reliably by grounding perception in explicit physical constraints.
Abstract
Freespace detection in autonomous driving is limited by the lack of explicit geometric modeling, hindering generalization across complex terrains. Existing approaches are predominantly data-driven and neglect the physical structure of drivable surfaces. We propose Terrain Flat (TerrFlat), a physics-driven geometric representation that models road sur- faces along three interpretable dimensions: lateral smoothness, longitudinal consistency, and vertical deviation. TerrFlat is constructed through geometric reasoning and projected into pixel-aligned maps via a differentiable projection, ensuring geometric–visual consistency. Building on this representation, we introduce a symmetric feature fusion module (SFFM) to integrate TerrFlat with visual features through bidirectional recalibration, improving semantic discrimination and boundary localization. Together, TerrFlat and SFFM form TerrFlat-Seg, a unified framework for physics-aware freespace perception. Experiments on KITTI-Road, Semantic-KITTI, and ORFD datasets demonstrate consistent improvements over existing baselines. Real-world validation on an automated guided vehicle platform further confirms the robustness of our approach.