Pith. sign in

REVIEW 5 major objections 6 minor 38 references

Satellite-derived traversability maps can supervise an image-based off-road navigation model, cutting real-world human interventions from eight to two while improving long-range heading accuracy.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-01 16:26 UTC pith:YEJ2YBH7

load-bearing objection A well-engineered method for distilling satellite-derived traversability priors into FPV heading affordances; the key numbers are softer than they look because the offline evaluation is self-referential, but the real-world deployment is genuinely useful evidence. the 5 major comments →

arxiv 2607.17984 v1 pith:YEJ2YBH7 submitted 2026-07-20 cs.RO

Distilling Global Traversability Priors for Image-based Affordance Prediction in Off-road Environments

Matthew Sivaprakasam , Samuel Triest , Micah Nye , Deegan Atha , Shehryar Khattak , David Fan , Wenshan Wang , Sebastian Scherer This is my paper
classification cs.RO
keywords off-road navigationaffordance predictionfrontier estimationsatellite traversability mapsheading-space supervisionlong-range autonomyself-supervised learningGaussian process regression
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper argues that the myopia of local-metric-map off-road navigation can be overcome by distilling global traversability knowledge into a heading-space affordance model that runs on first-person images. The authors build a robot-specific global traversability map from satellite imagery, using Gaussian process regression to propagate the robot's own local traversability experience onto unseen regions. Planning feasible long-range paths on that map yields frontier labels for every heading, replacing the need for extensive human demonstrations and avoiding the depth-estimation and point-tracking noise of prior image-space supervision. Training directly on the heading-space representation used by the planner produces fewer human interventions (2 vs 8 for a demonstration-driven baseline) in real-world long-range runs and a lower heading error in offline planning benchmarks. The paper's core wager is that an automatically generated satellite traversability map can serve as trustworthy supervision for terrain the robot has never visited.

Core claim

The central claim is that long-range off-road navigation can be improved by supervising an image-based affordance predictor in heading space with planner-derived labels from a global traversability map, instead of supervising heatmaps with human demonstrations or tracked future trajectories. The global map is built by registering self-supervised local traversability estimates into a satellite coordinate frame, regressing traversability over the whole map with a Gaussian process fitted on visual features extracted from satellite patches, and marking cliffs lethal from digital elevation data. Feasible paths to a diverse set of distant goals are computed with the Fast Marching Method, and each

What carries the argument

The load-bearing object is the heading-space frontier score A_k, decomposed as A_k = P_k P_k C*, where P_k is the probability that an optimal path lies in the k-th angular bin and C* is the traversability cost of that path; the model is supervised directly on this distribution with an Earth Mover's Distance loss. Its training signal is produced by a pipeline that converts satellite imagery plus the robot's local traversability experience into a global traversability map via Gaussian process regression on visual features (with steep slopes from digital elevation data marked lethal), then extracts feasible frontier headings using the Fast Marching Method to plan paths to diverse long-range goa

Load-bearing premise

The automatically generated satellite traversability map—calibrated only near the robot's trajectory and used to label every heading, including never-visited terrain—is accurate enough to serve as ground truth.

What would settle it

Run the full training pipeline in an area where the satellite map is demonstrably stale (e.g., after a recent landslide or vegetation clearing), compare the trained model's frontier heading error at the planning horizon against a demonstration-only baseline; if the satellite-supervised model is not better, the load-bearing assumption fails. Alternatively, compare the GP-regressed traversability predictions against fresh proprioceptive traversability measurements in never-visited regions; large discrepancies at long range would falsify the supervision signal.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Training in heading space still yields interpretable image-space affordance heatmaps, providing semantic visualization without image-level supervision.
  • Supplementing demonstrations with planner-generated frontier labels improves detection in multi-frontier 'fork' scenarios and in out-of-distribution environments with different cameras and terrain.
  • In real-world long-range courses, the method completes missions with fewer human interventions (2 total) and less total time (325.84 s) than a no-frontier baseline (16 interventions, 504.38 s) or a demonstration-driven baseline (8 interventions, 505.75 s).
  • The only additional runtime overhead is a lightweight decoder because the visual features are shared with the local traversability estimator.
  • The approach transfers to a different ATV platform with a wider field-of-view camera in offline evaluation, suggesting cross-platform generalization.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The method's ceiling is set by the satellite prior: as higher-resolution, more temporally current satellite imagery and DEMs become available, accuracy should scale without new robot data, since the supervision pipeline is fully automated.
  • The same 'distill a global prior into an image-space or heading-space module' recipe could be applied to other geometric or semantic global priors (seasonal vegetation, water depth, snow cover) given a differentiable planning oracle.
  • The paper notes the system has no memory of frontiers once out of view; accumulating predicted frontier headings over time into a graph or map would likely remove the remaining interventions that occur after the robot commits to a wrong turn.
  • A testable extension: use the global traversability map directly as a privileged teacher during deployment (when available) and measure the gap between privileged and image-only performance; the size of that gap would quantify how much global knowledge remains uncaptured.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

5 major / 6 minor

Summary. The paper proposes a method for learning long-range, heading-space affordance predictions from first-person-view (FPV) images in off-road navigation. Supervision is generated by converting satellite imagery into global traversability maps via SALON-registered Gaussian Process regression, planning Fast Marching Method paths to distant goals, and deriving frontier heading labels from those paths. The learned affordance model is then used to bias local MPPI planning. The authors compare against a no-frontier baseline, the Long Range Navigator (LRN), and a demonstration-only supervision (DOS) variant, reporting offline heading-error and Modified Hausdorff Distance metrics as well as real-world intervention counts on an ATV.

Significance. The core idea is attractive: supervising a heading-space affordance model with planning-derived labels from satellite traversability maps avoids the dependence on depth estimation and point-track noise of prior image-space methods, while still producing interpretable image heatmaps. The real-world experiments, although limited, are a commendable step beyond purely offline evaluation, and the OOD camera/platform test is a useful stress test. If the empirical claims were fully supported, this would be a meaningful contribution to long-horizon off-road navigation. At present, however, the evidence is not yet convincing: the offline evaluation is circular with respect to the same global map used to generate training labels, and all metrics lack error bars, confidence intervals, or significance tests. The paper's own conclusion concedes that satellite-derived supervision may be inaccurate, yet this central risk is not quantified.

major comments (5)
  1. [V-B.2 and IV-B] The offline evaluation is circular in an important sense. The 'privileged ground-truth path' in Section V-B.2 is planned over the full global traversability map generated in Section IV-B, and that same map is the source of the training labels for A_k. Thus the heading error (0.22 vs 0.28 rad) and MHD (3.76 vs 4.04) in Table II measure how well the model reproduces the FMM planner on the predicted satellite map, not whether those map-based judgments are correct in the real world. Please validate the GP-generated global map against held-out proprioceptive SALON maps or independent terrain labels, and report offline metrics on regions/courses whose traversability was not used for supervision.
  2. [V-C and Table III] The real-world evidence is a single pass over five courses with one aggregate intervention count per method. The headline 2 vs 8 interventions could be dominated by one lucky or unlucky decision. No per-course breakdown, error bars, or significance testing is provided, and the intervention criterion is only described verbally. Report per-course intervention counts and times, make the intervention criterion operational, and, if repeated runs are infeasible, provide a sensitivity analysis. As written, the non-circular evidence is too thin to support the central claim.
  3. [IV-B.2] The supervision for headings not on demonstrated trajectories relies entirely on Gaussian Process regression from DINOv2/JAFAR satellite features to traversability, calibrated only within the vehicle trajectory radius. The paper provides no quantitative validation of GP predictions in unvisited areas and does not use GP predictive variance, despite it being naturally available. Since the Conclusions concede that satellite imagery can differ substantially from robot experience, this is a load-bearing omission. Please add a held-out validation of GP-predicted traversability and/or mask low-confidence labels using predictive variance.
  4. [VI-B and Table II] The comparison between 'Ours' and 'DOS' is intended to isolate the effect of planner-generated label augmentation, but the loss function also changes: the proposed method uses EMD with the regularization term of Eq. (9), while DOS uses MSE plus a margin loss. The observed improvements in Table II and in real-world interventions could therefore be due to the loss/optimization rather than the planning-based augmentation. To support contribution (3), include a DOS variant trained with the same EMD-based loss on demonstration-only labels.
  5. [IV-A, Eq. (8)] The pixel-to-heading projection in Eq. (8) is described as relying only on known camera intrinsics. On an off-road ATV, pitch and roll are significant, and without attitude compensation a fixed camera's pixel columns do not correspond to fixed world headings on slopes. Please clarify whether IMU/attitude data is used to stabilize the heading projection during both training and deployment; if not, this is a correctness risk for the entire heading-space representation.
minor comments (6)
  1. [Table II] The table layout is hard to parse in the text: several values run together (e.g., '0.220.40', '8.93.765.5') and column separators are missing. Please reformat with clear columns and include entries for NFE on all rows.
  2. [Table I] Some entries appear to lack spacing (e.g., '0.460.69'). Check the table rendering and column alignment.
  3. [Abstract] The claim of 'more than 10%' improvement should specify the metric and whether it is relative or absolute; Table II shows about 21% relative improvement in heading error but only ~7% in MHD.
  4. [V-B.2] The paper states that goals are chosen using future demonstration data to guarantee feasibility, but does not clarify whether the held-out test trajectories also contributed to the global traversability map used for privileged planning and label generation. Please state this explicitly.
  5. [Eq. (8)] The case where a heading bin contains no pixels is not addressed. Specify how A_k is defined in that situation, since this can occur for occluded or out-of-view headings.
  6. [V-C] The definition of a human intervention is given in prose; please provide a more mechanical or checkable criterion, since intervention counts are the primary real-world metric.

Circularity Check

0 steps flagged

No significant circularity: the affordance model is supervised by planner-derived headings from a satellite traversability map, and the central real-world claim is supported by external intervention counts rather than by the model's own outputs.

full rationale

The derivation chain is not circular. Ground-truth headings A*_k are produced by an independent pipeline: SALON local traversability maps registered with RTK GPS, GP regression from DINOv2/JAFAR satellite features, and FMM planning to diverse goals (Section IV-B/C). The model output A_k is computed from an FPV image via Eq. 7-8, and the loss in Eq. 9 compares this output to A*_k; there is no equation where the model's prediction is fed back into its own label. SALON and TartanDrive 2.0 are self-citations, but SALON is described as an open-source package and TartanDrive 2.0 is a public dataset, so they are independent building blocks rather than unverified load-bearing citations. The offline heading-space benchmark does generate privileged paths from the same global traversability map that produced the training frontiers, and the paper explicitly hedges this with 'Assuming an accurate global traversability map'; this makes the offline numbers a distillation-consistency check rather than external validation of the map's real-world accuracy. That is a validity limitation, not a circular derivation. The paper's own conclusion concedes 'If the terrain in the satellite imagery differs greatly from what the robot experiences, the supervision signal is less likely to be accurate,' which is an explicit correctness caveat. The real-world experiments (2 interventions for the proposed method vs 8 for LRN and 16 for no-frontier baseline) provide an external outcome measure not derived from the same map, and the central claim therefore has independent empirical content.

Axiom & Free-Parameter Ledger

6 free parameters · 5 axioms · 0 invented entities

No new physical or latent entities are postulated. The global traversability map is an estimated quantity derived from data, not an invented construct with independent falsifiable consequences. The free parameters are the user-chosen hyperparameters of the pipeline; the axioms are the domain assumptions that must hold for the automatically generated labels to be correct.

free parameters (6)
  • K (number of heading bins)
    Number of angular bins in heading-space decomposition (eq. 4); not specified in the text, chosen by hand.
  • H (local planning horizon)
    Horizon in eq. (2)-(3) and deployment waypoint distance; not specified; determines the local/global split.
  • R (goal radius for FMM)
    Radius of goal circle for path generation in Section IV-C, given only as R >> H.
  • lambda_reg (regularization weight)
    Weight on L1 regularization in eq. (9); not specified.
  • Gaussian Process hyperparameters
    GP for global traversability regression in Section IV-B.2, fit to SALON-derived feature-traversability pairs; kernel and noise not reported.
  • D_k goal-conditioning variance
    Gaussian width used to compute D_k during deployment; borrowed from LRN but the parameter value is not reported.
axioms (5)
  • domain assumption Satellite RGB appearance is predictive of ground vehicle traversability outside the driven area
    Central to Section IV-B.2: GP regresses traversability on DINOv2/JAFAR features for pixels beyond SALON coverage; if this fails, labels are wrong.
  • domain assumption RTK GPS registration aligns local SALON maps to the satellite map without significant drift
    Required for the one-to-one mapping in Section IV-B.1.
  • standard math Fast Marching Method produces globally optimal shortest paths on the traversability map
    Section IV-C uses FMM [31] to generate feasible paths; standard algorithm on a grid.
  • domain assumption Occlusion can be modeled by 2D line-of-sight with no lethal terrain
    Section IV-C: frontier is the furthest point along the path not occluded by lethal terrain within the camera FOV; ignores elevation and vegetation occlusion.
  • domain assumption Earth Mover's Distance on cumulative heading distributions is a valid training signal for multi-modal frontier labels
    Section IV-D.1 eq. (9); no ablation compares EMD to other losses beyond the baseline comparison.

pith-pipeline@v1.3.0-alltime-deepseek · 10604 in / 9896 out tokens · 104093 ms · 2026-08-01T16:26:10.316763+00:00 · methodology

0 comments
read the original abstract

Standard methods for autonomous navigation in unstructured terrain are prone to myopic behaviors in long-horizon scenarios. The use of metric maps built from LiDAR or cameras provides necessary local geometry and semantic information but is strictly limited by depth sensing range. By discarding data beyond the mapping horizon robots suffer from suboptimal, short-sighted decisions. To recover this lost information, we focus on extracting long-range traversability-aware frontiers directly from first-person-view (FPV) images. By leveraging satellite imagery, we compute the set of feasible navigation paths for a dataset of image/pose pairs and use them to supervise our network, reducing the need for extensive human demonstration data. We demonstrate that this approach improves performance in long-range off-road navigation over existing methods by more than 10% in various offline benchmarks and reduces the number of human interventions incurred in a set of real-world experiments. More details can be found at https://theairlab.org/ss_frontiers_iros .

Figures

Figures reproduced from arXiv: 2607.17984 by David Fan, Deegan Atha, Matthew Sivaprakasam, Micah Nye, Samuel Triest, Sebastian Scherer, Shehryar Khattak, Wenshan Wang.

Figure 1
Figure 1. Figure 1: Overview of our training pipeline. FPV images are fed into a VFM, the output of which in turn is fed into a decoder that produces an affordance [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Visualization of the components Pk and C∗ needed to compute Vk via eq. 4-6. Given a set of optimal long-range paths, C∗ models their traversability costs and Pk models its probability of existence. compute Vk for a set of K angular bins with heading θk around the robot. Vk is decomposed as: Vk(x1, xG, xH) = Ak(x1, xH)Dk(xH, xG) (4) V (xH) = Vk∗ , k∗ = arg min k d(xH, θk) (5) where d computes the angular di… view at source ↗
Figure 3
Figure 3. Figure 3: Global traversability map computation: We start with an off-road driving dataset, used to generate local traversability maps at each timestep and [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Instead of relying solely on demonstration (shown in dashed white), [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 1
Figure 1. Figure 1: We convert both into categorical distributions, [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative comparison of our method to LRN. Noisy point tracks and terrain occlusions cause inaccurate training signals for LRN, resulting [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Resulting paths taken by our method and baselines in a subset of our real-world experiments. Our method incurs the fewest total interventions [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

38 extracted references · 2 canonical work pages

  1. [1]

    NoMaD: Goal Masked Diffusion Policies for Navigation and Exploration,

    A. Sridhar, D. Shah, C. Glossop, and S. Levine, “NoMaD: Goal Masked Diffusion Policies for Navigation and Exploration,”arXiv pre-print, 2023. [Online]. Available: https://arxiv.org/abs/2310.07896

    Pith/arXiv arXiv 2023

  2. [2]

    Risk-guided diffusion: Toward deploying robot foundation models in space, where failure is not an option,

    R. Thakker, A. Patnaik, V . Kurtz, J. Frey, J. Becktor, S. Moon, R. Royce, M. Kaufmann, G. Georgakis, P. Roth, J. W. Burdick, M. Hutter, and S. Khattak, “Risk-guided diffusion: Toward deploying robot foundation models in space, where failure is not an option,” inRSS 2025 Workshop on Reliable Robotics: Safety and Security in the Face of Generative AI, 2025...

    2025

  3. [3]

    Long range navigator (LRN): Extending robot planning horizons beyond metric maps,

    M. Schmittle, R. Baijal, N. Hatch, R. Scalise, M. G. Castro, S. Talia, K. Khetarpal, B. Boots, and S. Srinivasa, “Long range navigator (LRN): Extending robot planning horizons beyond metric maps,” in 9th Annual Conference on Robot Learning, 2025. [Online]. Available: https://openreview.net/forum?id=QtVZUPCKrY

    2025

  4. [4]

    Vamos: A hierarchical vision-language-action model for capability-modulated and steerable navigation,

    M. G. Castro, S. Rajagopal, D. Gorbatov, M. Schmittle, R. Baijal, O. Zhang, R. Scalise, S. Talia, E. Romig, C. de Melo, B. Boots, and A. Gupta, “Vamos: A hierarchical vision-language-action model for capability-modulated and steerable navigation,”Under Review, 2025

    2025

  5. [5]

    Learning to move with affordance maps,

    W. Qi, R. T. Mullapudi, S. Gupta, and D. Ramanan, “Learning to move with affordance maps,” inInternational Conference on Learning Representations, 2020. [Online]. Available: https: //openreview.net/forum?id=BJgMFxrYPB

    2020

  6. [6]

    Raven: Resilient aerial navigation via open-set semantic memory and behavior adaptation,

    S. Kim, O. Alama, D. Kurdydyk, J. Keller, N. Keetha, W. Wang, Y . Bisk, and S. Scherer, “Raven: Resilient aerial navigation via open-set semantic memory and behavior adaptation,”arXiv preprint arXiv:2509.23563, 2025

    arXiv 2025

  7. [7]

    Salon: Self-supervised adaptive learning for off-road navigation,

    M. Sivaprakasam, S. Triest, C. Ho, S. Aich, J. Lew, I. Adu, W. Wang, and S. Scherer, “Salon: Self-supervised adaptive learning for off-road navigation,” in2025 IEEE International Conference on Robotics and Automation (ICRA), 2025, pp. 16 999–17 006

    2025

  8. [8]

    Terrainnet: Visual modeling of complex terrain for high-speed, off-road navigation,

    X. Meng, N. Hatch, A. Lambert, A. Li, N. Wagener, M. Schmittle, J. Lee, W. Yuan, Z. Chen, S. Deng, G. Okopal, D. Fox, B. Boots, and A. Shaban, “Terrainnet: Visual modeling of complex terrain for high-speed, off-road navigation,” 2023

    2023

  9. [9]

    Step: Stochastic traversability evaluation and planning for risk-aware off-road navigation; results from the darpa subterranean challenge,

    A. Dixit, D. Fan, K. Otsu, S. Dey, A.-A. Agha-Mohammadi, and J. Burdick, “Step: Stochastic traversability evaluation and planning for risk-aware off-road navigation; results from the darpa subterranean challenge,”Field Robotics, vol. 4, no. 1, p. 182–210, Jan. 2024. [Online]. Available: http://dx.doi.org/10.55417/fr.2024006

    work page doi:10.55417/fr.2024006 2024

  10. [10]

    Velociraptor: Leveraging visual foundation models for label-free, risk-aware off-road navigation,

    S. Triest, M. Sivaprakasam, S. Aich, D. Fan, W. Wang, and S. Scherer, “Velociraptor: Leveraging visual foundation models for label-free, risk-aware off-road navigation,” in8th Annual Conference on Robot Learning, 2024. [Online]. Available: https: //openreview.net/forum?id=AhEE5wrcLU

    2024

  11. [11]

    Creste: Scalable map- less navigation with internet scale priors and counterfactual guidance,

    A. Zhang, H. Sikchi, A. Zhang, and J. Biswas, “Creste: Scalable map- less navigation with internet scale priors and counterfactual guidance,” inRobotics: Science and Systems (RSS), 2025

    2025

  12. [12]

    Wild visual navigation: Fast traversability learning via pre-trained models and online self- supervision,

    M. Mattamala, J. Frey, P. Libera, N. Chebrolu, G. Martius, C. Cadena, M. Hutter, and M. Fallon, “Wild visual navigation: Fast traversability learning via pre-trained models and online self- supervision,”Autonomous Robots, vol. 49, no. 3, p. 19. [Online]. Available: https://doi.org/10.1007/s10514-025-10202-x

    doi:10.1007/s10514-025-10202-x

  13. [13]

    How does it feel? self-supervised costmap learning for off-road vehicle traversability,

    M. G. Castro, S. Triest, W. Wang, J. M. Gregory, F. Sanchez, J. G. Rogers, and S. Scherer, “How does it feel? self-supervised costmap learning for off-road vehicle traversability,” in2023 IEEE International Conference on Robotics and Automation (ICRA), 2023, pp. 931–938

    2023

  14. [14]

    The value of planning for infinite-horizon model predictive control,

    N. Hatch and B. Boots, “The value of planning for infinite-horizon model predictive control,” in2021 IEEE International Conference on Robotics and Automation (ICRA). IEEE Press, 2021, p. 7372–7378. [Online]. Available: https://doi.org/10.1109/ICRA48506. 2021.9561718

    arXiv 2021

  15. [15]

    Mapexrl: Human-inspired indoor exploration with predicted en- vironment context and reinforcement learning,

    N. Harutyunyan, B. Moon, S. Kim, C. Ho, A. Hung, and S. Scherer, “Mapexrl: Human-inspired indoor exploration with predicted en- vironment context and reinforcement learning,”arXiv preprint arXiv:2503.01548, 2025

    arXiv 2025

  16. [16]

    Dare: Diffusion policy for autonomous robot exploration,

    Y . Cao, J. Lew, J. Liang, J. Cheng, and G. Sartoretti, “Dare: Diffusion policy for autonomous robot exploration,” in2025 IEEE International Conference on Robotics and Automation (ICRA), 2025, pp. 11 987– 11 993

    2025

  17. [17]

    Learning over subgoals for efficient navigation of structured, unknown environments,

    G. J. Stein, C. Bradley, and N. Roy, “Learning over subgoals for efficient navigation of structured, unknown environments,” in Proceedings of The 2nd Conference on Robot Learning, ser. Proceedings of Machine Learning Research, A. Billard, A. Dragan, J. Peters, and J. Morimoto, Eds., vol. 87. PMLR, 29–31 Oct 2018, pp. 213–222. [Online]. Available: https://...

    2018

  18. [18]

    Omnivla: An omni-modal vision-language-action model for robot navigation,

    N. Hirose, C. Glossop, D. Shah, and S. Levine, “Omnivla: An omni-modal vision-language-action model for robot navigation,”

  19. [19]

    Ventura: Adapting image diffusion models for unified task conditioned navigation,

    A. Zhang, X. Meng, L. Calliari, D.-K. Kim, S. Omidshafiei, J. Biswas, A. Agha, and A. Shaban, “Ventura: Adapting image diffusion models for unified task conditioned navigation,” inIEEE International Con- ference on Robotics and Automation (ICRA), 2026

    2026

  20. [20]

    Learning long- range vision for autonomous off-road driving,

    R. Hadsell, P. Sermanet, J. Ben, A. Erkan, M. Scoffier, K. Kavukcuoglu, U. Muller, and Y . LeCun, “Learning long- range vision for autonomous off-road driving,”Journal of Field Robotics, vol. 26, no. 2, pp. 120–144, 2009. [Online]. Available: https://onlinelibrary.wiley.com/doi/abs/10.1002/rob.20276

    work page doi:10.1002/rob.20276 2009

  21. [21]

    Far-field image-based traversability mapping for a priori unknown natural environments,

    E. Fahnestock, E. Fuentes, S. Prentice, V . Vasilopoulos, P. R. Osteen, T. Howard, and N. Roy, “Far-field image-based traversability mapping for a priori unknown natural environments,”IEEE Robotics and Automation Letters, vol. 10, no. 6, pp. 6039–6046, 2025

    2025

  22. [22]

    CoTracker: It is better to track together,

    N. Karaev, I. Rocco, B. Graham, N. Neverova, A. Vedaldi, and C. Rupprecht, “CoTracker: It is better to track together,” 2023

    2023

  23. [23]

    Learning to search: Functional gradient techniques for imitation learning,

    N. Ratliff, D. Silver, and J. A. D. Bagnell, “Learning to search: Functional gradient techniques for imitation learning,”Autonomous Robots, vol. 27, no. 1, pp. 25 – 53, July 2009

    2009

  24. [24]

    Pacer: Preference- conditioned all-terrain costmap generation,

    L. Mao, G. Warnell, P. Stone, and J. Biswas, “Pacer: Preference- conditioned all-terrain costmap generation,”Robotics and Automation Letters, 2025

    2025

  25. [25]

    Generating robot-dependent cost maps for off-road environments using locomo- tion experiments and earth observation data,

    M. Eder, R. Prinz, F. Sch ¨oggl, and G. Steinbauer-Wagner, “Generating robot-dependent cost maps for off-road environments using locomo- tion experiments and earth observation data,” in2022 Sixth IEEE International Conference on Robotic Computing (IRC), 2022, pp. 172– 176

    2022

  26. [26]

    Vlm-gronav: Robot navi- gation using physically grounded vision-language models in outdoor environments,

    M. Elnoor, K. Weerakoon, G. Seneviratne, R. Xian, T. Guan, M. K. M Jaffar, V . Rajagopal, and D. Manocha, “Vlm-gronav: Robot navi- gation using physically grounded vision-language models in outdoor environments,” in2025 IEEE International Conference on Robotics and Automation (ICRA), 2025, pp. 2391–2398

    2025

  27. [27]

    Map it anywhere (mia): Empowering bird’s eye view mapping using large-scale public data,

    C. Ho, J. Zou, O. Alama, S. M. J. Kumar, B. Chiang, T. Gupta, C. Wang, N. Keetha, K. Sycara, and S. Scherer, “Map it anywhere (mia): Empowering bird’s eye view mapping using large-scale public data,” inAdvances in Neural Information Processing Systems, 2024. [Online]. Available: https://arxiv.org/abs/2407.08726

    Pith/arXiv arXiv 2024

  28. [28]

    Roadrunner m&m - learning multi-range multi-resolution traversabil- ity maps for autonomous off-road navigation,

    M. Patel, J. Frey, D. Atha, P. Spieler, M. Hutter, and S. Khattak, “Roadrunner m&m - learning multi-range multi-resolution traversabil- ity maps for autonomous off-road navigation,”IEEE Robotics and Automation Letters, vol. 9, no. 12, pp. 11 425–11 432, 2024

    2024

  29. [29]

    Dinov2: Learning robust visual features without supervision,

    M. Oquab, T. Darcet, T. Moutakanni, H. V . V o, M. Szafraniec, V . Khalidov, P. Fernandez, D. Haziza, F. Massa, A. El-Nouby, R. Howes, P.-Y . Huang, H. Xu, V . Sharma, S.-W. Li, W. Galuba, M. Rabbat, M. Assran, N. Ballas, G. Synnaeve, I. Misra, H. Jegou, J. Mairal, P. Labatut, A. Joulin, and P. Bojanowski, “Dinov2: Learning robust visual features without ...

    2023

  30. [30]

    Jafar: Jack up any feature at any resolution,

    P. Couairon, L. Chambon, L. Serrano, J.-E. Haugeard, M. Cord, and N. Thome, “Jafar: Jack up any feature at any resolution,” 2025. [Online]. Available: https://arxiv.org/abs/2506.11136

    arXiv 2025

  31. [31]

    A fast marching level set method for monotonically advancing fronts

    J. A. Sethian, “A fast marching level set method for monotonically advancing fronts.”Proceedings of the National Academy of Sciences, vol. 93, no. 4, pp. 1591–1595, 1996. [Online]. Available: https://www.pnas.org/doi/abs/10.1073/pnas.93.4.1591

    doi:10.1073/pnas.93.4.1591 1996

  32. [32]

    Am-radio: Agglomerative vision foundation model reduce all domains into one,

    M. Ranzinger, G. Heinrich, J. Kautz, and P. Molchanov, “Am-radio: Agglomerative vision foundation model reduce all domains into one,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2024, pp. 12 490–12 500

    2024

  33. [33]

    EVORA: Deep Evidential Traversability Learning for Risk-Aware Off-Road Autonomy,

    X. Cai, S. Ancha, L. Sharma, P. R. Osteen, B. Bucher, S. Phillips, J. Wang, M. Everett, N. Roy, and J. P. How, “EVORA: Deep Evidential Traversability Learning for Risk-Aware Off-Road Autonomy,”IEEE Transactions on Robotics, vol. 40, pp. 3756–3777, 2024

    2024

  34. [34]

    Squared earth mover’s distance-based loss for training deep neural networks,

    L. Hou and C.-P. Yu, “Squared earth mover’s distance-based loss for training deep neural networks,” 11 2016

    2016

  35. [35]

    Tartandrive 2.0: More modalities and better infrastructure to further self-supervised learning research in off-road driving tasks,

    M. Sivaprakasam, P. Maheshwari, M. G. Castro, S. Triest, M. Nye, S. Willits, A. Saba, W. Wang, and S. Scherer, “Tartandrive 2.0: More modalities and better infrastructure to further self-supervised learning research in off-road driving tasks,” in2024 IEEE International Conference on Robotics and Automation (ICRA), 2024, pp. 12 606– 12 606

    2024

  36. [36]

    Model predictive path integral control: From theory to parallel computation,

    G. Williams, A. Aldrich, and E. Theodorou, “Model predictive path integral control: From theory to parallel computation,”Journal of Guidance, Control, and Dynamics, vol. 40, pp. 1–14, 01 2017

    2017

  37. [37]

    Rayfronts: Open-set semantic ray frontiers for online scene understanding and exploration,

    O. Alama, A. Bhattacharya, H. He, S. Kim, Y . Qiu, W. Wang, C. Ho, N. Keetha, and S. Scherer, “Rayfronts: Open-set semantic ray frontiers for online scene understanding and exploration,” in2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2025, pp. 5930–5937

    2025

  38. [2025]

    Available: https://arxiv.org/abs/2509.19480

    [Online]. Available: https://arxiv.org/abs/2509.19480

    arXiv