DGT-Map: Directional Global Traversability Mapping Utilizing Multi-Task Learning for Heterogeneous Vehicles

Summary

Off-road traversability is direction-dependent and vehicle specific, yet most global maps assign a single isotropic cost to each location.

Abstract Summary

Off-road traversability is direction-dependent and vehicle specific, yet most global maps assign a single isotropic cost to each location. Existing learned estimators are also commonly trained independently for each vehicle; this preserves vehicle-specific behavior but prevents vehicles from sharing common terrain representations. DGT-MAP addresses both limitations through a self-supervised framework that learns global, directional, and vehicle-conditioned traversability costmaps from RGB-D observations and locomotion signals. A shared multi-task backbone learns common terrain features across training vehicles while vehicle-specific prediction heads preserve platform-dependent responses. At inference, DGT-MAP produces a heading-indexed costmap that can be used by a direction-aware planner. We evaluate DGT-MAP in simulation by integrating it into a Hybrid A* navigation stack and measuring downstream task success on challenging terrains, including slopes that are traversable downhill but not uphill and a ridge obstacle that is traversable by some vehicles, but not by others. Across evaluated tasks, DGT-MAP achieves the highest or tied-highest navigation success rate when compared against geometric, binary, and learned direction-agnostic baselines.

Key Points

Off-road traversability is direction-dependent and vehicle specific, yet most global maps assign a single isotropic cost to each location.
Existing learned estimators are also commonly trained independently for each vehicle; this preserves vehicle-specific behavior but prevents vehicles from sharing common terrain representations.
DGT-MAP addresses both limitations through a self-supervised framework that learns global, directional, and vehicle-conditioned traversability costmaps from RGB-D observations and locomotion signals.
A shared multi-task backbone learns common terrain features across training vehicles while vehicle-specific prediction heads preserve platform-dependent responses.
At inference, DGT-MAP produces a heading-indexed costmap that can be used by a direction-aware planner.

Problem Statement

We consider the problem of learning a global, directional, vehicle-conditioned traversability map for off-road navigation using self-supervised data. Given $N$ vehicles equipped with RGB-D cameras and IMUs operating in an unknown off-road environment, our goal is to estimate terrain-induced traversal cost as a function of spatial location, direction of travel, and vehicle platform.

Methodology

Our approach learns the traversability function defined in Sec. III through self-supervision: rather than manually labeling terrain, we use the discrepancy between commanded and actual vehicle motion as a natural measure of traversal difficulty ( Sec. IV-A ).

We then use a multi-task CNN with a shared terrain backbone and vehicle-specific prediction heads, allowing terrain features to be shared across vehicles while preserving vehicle-dependent responses ( Sec. IV-B ).

Methodology — DGT-Map: Directional Global Traversability Mapping Utilizing Multi-Task Learning for Heterogeneous Vehicles — Fig. 3 : Traversability-aware planning. (a) When expanding from grid cell $(i,j)$, the planner computes the travel direction to neighboring cells $(i^{\prime}_{n},j^{\prime}_{n})$ for $n={1,2,3,4}$ and queries the corresponding directional cost layer. (b) Each candidate motion primitive is discretized and checked against the occupancy map to ensure collision-free feasibility. (c) Four of the $K=8$ directional costmap layers ($\theta=0^{\circ},90^{\circ},180^{\circ},270^{\circ}$) illustrating how traversability varies with heading; differences across layers capture anisotropic terrain effects such as ascending versus descending slopes.

Experiments

In this section, we empirically evaluate four central claims of our method: dgt-map captures global, directional traversability information not captured by direction-agnostic methods. dgt-map captures vehicle-specific nuances in traversability. dgt-map captures features beyond basic geometric slope cues.

Multi-task training improves performance compared to single-task training with the same amount of vehicle-specific data, especially in cases where vehicle-specific data is limited.

Experiments — DGT-Map: Directional Global Traversability Mapping Utilizing Multi-Task Learning for Heterogeneous Vehicles — Fig. 4 : Top Left: Hill Map showing Route A, a roundabout route with a gentle slope, and Route B, a direct route containing a steep section. Top Right: Ridge Map showing Route A, a path that avoids the sharp ridge, and Route B, a path that crosses the ridge. Route B is not traversable by some vehicles. Bottom Left: Traversability map of the Jeep , note how the ridge is shown as traversable. Bottom Right: Traversability map for the MercedesVan , note how the ridge has a low traversability.

Sources and Demonstrations

Figures are reproduced from Singh et al. See the full paper for experimental details and the project page, when the paper links one, for demonstrations and videos.