Search papers, labs, and topics across Lattice.
This paper introduces a novel approach for learning traversability-aware global planners that leverage overhead geo-spatial data to enhance long-horizon off-road navigation. By utilizing human-driven GPS trajectories and self-supervised geometric priors from LiDAR, the proposed method effectively creates a continuous traversability map, addressing the limitations of existing short-range sensing techniques. In practical field trials, the model demonstrated impressive performance, achieving trajectories within 3.66% of human path length while significantly reducing operator interventions by approximately 85%.
Learning traversability from overhead data can cut operator interventions by 85% while matching human navigation accuracy in challenging off-road environments.
Autonomous navigation across large off-road environments remains a challenging problem. Onboard sensors perceive only the immediate surroundings, yet safe and efficient routes depend on terrain features that extend well beyond the sensor horizon. Geo-spatial data sources such as satellite imagery, aerial LiDAR, and vector maps can close this gap, but learning traversability from them is difficult: dense labels are unavailable at scale, and existing methods rely on short-range sensing. We propose an efficient formulation that learns a continuous traversability map from overhead data, supervised directly by human-driven GPS trajectories and shaped by self-supervised geometric priors from LiDAR. Alongside the model, we release a public dataset of 299 scenes spanning $\sim\!1{,}244\,\mathrm{km}^{2}$ of diverse terrain, paired with $1{,}130\,\mathrm{km}$ of human driving. In field trials on a Clearpath Warthog across seven routes at two sites, our method achieves trajectories within $3.66\%$ of human path length and reduces operator interventions by $\sim\!85\%$ compared to local-planner-only autonomy.
