TADPO: Reinforcement Learning Goes Off-road

While Vision-Language Models (VLMs) enable high-level semantic reasoning for end-to-end autonomous driving, particularly in unstructured environments, existing off-road datasets suffer from language annotations that are weakly aligned with vehicle actions and terrain geometry. To address this misalignment, we propose a language refinement framework that restructures annotations into action-aligned pairs, enabling a VLM to generate refined scene descriptions and 3D future trajectories directly from a single image. To further encourage terrain-aware planning, we introduce a preference optimization strategy that constructs geometry-aware hard negatives and explicitly penalizes trajectories inconsistent with local elevation profiles. Furthermore, we propose off-road-specific metrics to quantify traversability compliance and elevation consistency, addressing the limitations of conventional on-road evaluation. Experiments on the ORAD-3D benchmark demonstrate that our approach reduces average trajectory error from 1.01m to 0.97m, improves traversability compliance from 0.621 to 0.644, and decreases elevation inconsistency from 0.428 to 0.322, highlighting the efficacy of action-aligned supervision and terrain-aware optimization for robust off-road driving.

Human demonstrations are widely considered the cornerstone of end-to-end (E2E) autonomous driving despite human demonstration's scarcity for long-tail and safety-critical scenarios. Nonetheless, current E2E autonomous driving (AD) training paradigms continue to rely on human demonstrations. Imitation learning (IL) requires human demonstrations for training, whereas reinforcement learning (RL) has emerged as a promising alternative to reduce this dependency. However, most existing RL methods for E2E AD still rely implicitly on human demonstrations. A pure rewards-based RL method can overcome the need for human demonstrations, but general RL policy gradient methods suffer from the cold-start problem. In this paper, we propose ZTRS (Zero-human demonstration end-to-end autonomous driving with TRajectory Scorer) - a complete RL-based E2E planning paradigm trained solely on real-world images and rule-based rewards, entirely without human demonstration. Through our proposed Exhaustive Policy Optimization (EPO), a policy gradient variant tailored for enumerable trajectory actions and dense supervision, ZTRS enables the model to generalize better to long-tail driving scenarios. We demonstrate this generalization through our SOTA performance against IL approaches on both long-tail Navhard and closed-loop HUGSIM datasets. Project page: https://zhenxinli.net/ZTRS/.

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 supervised geometric priors from LiDAR. Alongside the model, we release a dataset, curated from public sources, consisting 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 $5.5\%$ of human path length and reduces operator interventions by $\sim\!85\%$ compared to local-planner-only autonomy.