Event
DGT-Map: Heterogeneous Vehicle Directional Traversability Map
1 reports · 1 sources
Jaskrit Singh and other authors released DGT-Map, a self-supervised framework that learns a global, directional, and vehicle-conditioned traversability cost map from RGB-D observations and motion signals. Its shared multi-task backbone learns terrain features common across vehicles, while vehicle-specific prediction heads preserve platform-dependent responses; at inference it outputs a heading-indexed cost map for use by direction-aware planners. The authors report that in simulation they integrated DGT-Map into a Hybrid A* navigation stack and evaluated it by downstream task success rate, on test terrains including slopes that can only be descended but not ascended, and ridge-like obstacles that some vehicles can pass but others cannot.
Report timeline
- DGT-Map: Directional Global Traversability Mapping with Multi-Task Learning for Heterogeneous Vehicles
DGT-Map proposes a self-supervised framework that learns global, directional, and vehicle-conditioned traversability costmaps from RGB-D observations and locomotion signals. It uses a shared multi-task backbone to learn common terrain features across vehicles, while vehicle-specific prediction heads preserve platform-dependent responses. At inference, it outputs a heading-indexed costmap for use by direction-aware planners.
arXiv cs.RO ·