pith. sign in

arxiv: 2604.19962 · v2 · submitted 2026-04-21 · 💻 cs.RO

Radar Odometry Subject to High Tilt Dynamics of Subarctic Environments

Mat\v{e}j Boxan , William Larriv\'ee-Hardy , Fran\c{c}ois Pomerleau This is my paper

Pith reviewed 2026-05-10 01:48 UTC · model grok-4.3

classification 💻 cs.RO
keywords radar odometryinertial odometrysubarctic terraintilt dynamicsFMCW radarmotion estimationstate estimationoff-road navigation
0
0 comments X

The pith

A radar-inertial odometry method using tilt-proximity submap search and vertical displacement thresholds maintains accuracy under high pitch and roll in subarctic terrain.

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

The paper shows that conventional rotating FMCW radar odometry methods lose reliability when consecutive scans differ by up to 13 degrees in pitch and absolute angles reach 30 degrees pitch and 8 degrees roll, conditions common in subarctic settings. It introduces a new approach that restricts submap lookup to regions of similar tilt and applies a fixed cutoff to remove scan points whose vertical offset exceeds the estimated rotation axis. On an urban reference dataset the method matches the best prior results, while on a 2 km dynamic subarctic trajectory it records a 0.3 percent accuracy gain over the next strongest comparator. The work additionally dissects the failure patterns of all four techniques when the platform experiences large lateral slip and crosses steep ditches.

Core claim

We benchmark three existing radar odometry pipelines under demanding subarctic tilt conditions and present a radar-inertial method that searches submaps according to tilt proximity while discarding points violating a hard vertical-displacement threshold relative to the rotation axis. This yields state-of-the-art performance on urban baselines and a 0.3 percent improvement on a 2-kilometer trajectory with pronounced dynamics, together with a detailed breakdown of remaining error sources during high-slip and ditch-crossing maneuvers.

What carries the argument

tilt-proximity submap search paired with a hard threshold on vertical displacement between scan points and the estimated axis of rotation

If this is right

  • Radar odometry becomes usable for state estimation on non-flat subarctic ground without requiring an explicit flat-ground assumption.
  • A modest accuracy edge on long dynamic paths supports longer autonomous traverses in rough terrain.
  • Explicit analysis of slip and ditch failures identifies concrete remaining limits that any future method must address.
  • The benchmark data set provides a reference for comparing new tilt-aware odometry designs.

Where Pith is reading between the lines

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

  • The same tilt-proximity logic could be transferred to lidar or camera-based odometry in comparable off-road settings.
  • Larger gains may appear if the threshold mechanism is combined with online slip estimation rather than treated as a static filter.
  • The reported 0.3 percent margin would benefit from repeated trials and error statistics to confirm it exceeds run-to-run variability.
  • Successful tilt handling opens the possibility of reliable mapping loops in GPS-denied arctic regions where terrain changes rapidly.

Load-bearing premise

The specific hard threshold on vertical displacement and the tilt-proximity submap search remain effective outside the single 2 km test trajectory examined.

What would settle it

Re-running the four methods on additional subarctic trajectories that exhibit comparable tilt ranges and checking whether the 0.3 percent improvement persists or falls inside the range of normal variation.

Figures

Figures reproduced from arXiv: 2604.19962 by Fran\c{c}ois Pomerleau, Mat\v{e}j Boxan, William Larriv\'ee-Hardy.

Figure 1
Figure 1. Figure 1: Typical rotating FMCW radar solutions operate in flat terrains, [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: A diagram of our method’s pipeline. Bottom images show different stages of the processed radar data, where the border colors match the [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The Uncrewed Ground Vehicle equipped with a Navtech CIR-304 [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Comparison of the performance of the four tested methods on two [PITH_FULL_IMAGE:figures/full_fig_p003_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: Detailed view on the estimated trajectory from the four evaluated [PITH_FULL_IMAGE:figures/full_fig_p004_6.png] view at source ↗
read the original abstract

Rotating FMCW radar odometry methods often assume flat ground conditions. While this assumption is sufficient in many scenarios, including urban environments or flat mining setups, the highly dynamic terrain of subarctic environments poses a challenge to standard feature extraction and state estimation techniques. This paper benchmarks three existing radar odometry methods under demanding conditions, exhibiting up to 13{\deg} in pitch and 4{\deg} in roll difference between consecutive scans, with absolute pitch and roll reaching 30{\deg} and 8{\deg}, respectively. Furthermore, we propose a novel radar-inertial odometry method utilizing tilt-proximity submap search and a hard threshold for vertical displacement between scan points and the estimated axis of rotation. Experimental results demonstrate a state-of-the-art performance of our method on an urban baseline and a 0.3% improvement over the second-best comparative method on a 2-kilometer-long dynamic trajectory. Finally, we analyze the performance of the four evaluated methods on a complex radar sequence characterized by high lateral slip and a steep ditch traversal.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript benchmarks three existing radar odometry methods on trajectories with high tilt dynamics (up to 13° pitch/roll differences between scans) and proposes a novel radar-inertial odometry approach that augments standard feature extraction with a tilt-proximity submap search and a hard threshold on vertical displacement between scan points and the estimated rotation axis. It reports state-of-the-art performance on an urban baseline dataset and a 0.3% improvement over the second-best comparator on a single 2 km subarctic trajectory, plus qualitative analysis on a high-slip ditch sequence.

Significance. If the reported gains prove robust, the work would provide a targeted, low-overhead adaptation for radar-based localization in non-flat, high-tilt environments that are currently underserved by flat-ground assumptions. The benchmarking component supplies useful comparative data for the community. However, the 0.3% margin on a single trajectory without statistical support limits the immediate impact; stronger validation would be needed before the method could be considered a reliable advance.

major comments (2)
  1. [Results / Experimental Evaluation] Results section (and abstract): the headline claim of a 0.3% improvement over the second-best method on the 2 km dynamic trajectory is presented as a point estimate with no standard deviation, no repeated runs, and no sensitivity analysis on the hard vertical-displacement threshold or tilt-proximity radius. Given that both parameters are free and sequence-specific, this margin cannot be distinguished from run-to-run variation or tuning effects.
  2. [Proposed Method] Method description: the hard threshold on vertical displacement and the tilt-proximity submap search are introduced without derivation or ablation; no justification is given for why these particular heuristics remain effective outside the 2 km test sequence, nor is their interaction with the inertial component quantified.
minor comments (2)
  1. [Abstract] The abstract and results would benefit from explicit statement of the number of independent trials and the exact metric (e.g., ATE, RPE) used for the 0.3% figure.
  2. [Figures] Figure captions for the trajectory plots should include the quantitative error values rather than relying solely on qualitative description.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below and indicate the revisions planned for the next manuscript version.

read point-by-point responses

  1. Referee: [Results / Experimental Evaluation] Results section (and abstract): the headline claim of a 0.3% improvement over the second-best method on the 2 km dynamic trajectory is presented as a point estimate with no standard deviation, no repeated runs, and no sensitivity analysis on the hard vertical-displacement threshold or tilt-proximity radius. Given that both parameters are free and sequence-specific, this margin cannot be distinguished from run-to-run variation or tuning effects.

    Authors: We agree that the 0.3% figure is a single point estimate on one trajectory and that sensitivity analysis is warranted. In the revised manuscript we will qualify the claim in the abstract and results, explicitly noting the single-sequence nature of the evaluation. We will add a sensitivity study varying the tilt-proximity radius and vertical-displacement threshold across reasonable ranges and report the resulting error variation. The pipeline is deterministic for a fixed input sequence and parameter set, so repeated-run standard deviation does not apply; however, we will discuss parameter sensitivity as a proxy for robustness. revision: yes

  2. Referee: [Proposed Method] Method description: the hard threshold on vertical displacement and the tilt-proximity submap search are introduced without derivation or ablation; no justification is given for why these particular heuristics remain effective outside the 2 km test sequence, nor is their interaction with the inertial component quantified.

    Authors: The tilt-proximity submap search and vertical-displacement threshold are motivated by the observed inter-scan pitch/roll differences of up to 13° and the geometric requirement that scan points remain consistent with the IMU-derived rotation axis under high tilt. In the revision we will add a short geometric derivation subsection explaining these choices and include ablation experiments that isolate each component (with and without tilt-proximity search, with and without the vertical threshold) on both the urban and subarctic sequences. We will also quantify the interaction with the inertial estimator by comparing against a pure radar variant. revision: yes

Circularity Check

0 steps flagged

No circularity: experimental method with no load-bearing derivations or self-referential fits

full rationale

The paper describes an algorithmic extension to radar-inertial odometry (tilt-proximity submap search plus a hard vertical-displacement threshold) and reports empirical results on held-out trajectories. No equations, uniqueness theorems, or parameter-fitting steps are shown that reduce by construction to the inputs or to prior self-citations. The 0.3 % improvement is presented as an experimental outcome rather than a derived prediction, so the derivation chain contains no self-definitional, fitted-input, or self-citation-load-bearing reductions.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 0 invented entities

The method relies on standard rigid-body motion assumptions plus two new heuristics whose exact parameter values are not stated in the abstract; no new physical entities are postulated.

free parameters (2)
  • hard threshold for vertical displacement
    A fixed cutoff is applied to discard points; its value must be chosen or tuned to the sensor and environment.
  • tilt-proximity search radius or tolerance
    Parameters controlling which submaps are considered based on tilt similarity are required but not quantified in the abstract.
axioms (1)
  • domain assumption Radar point clouds can be aligned under rigid-body motion once tilt is accounted for by submap selection and vertical thresholding.
    Invoked when describing the novel method and its comparison to flat-ground baselines.

pith-pipeline@v0.9.0 · 5494 in / 1435 out tokens · 45461 ms · 2026-05-10T01:48:33.280267+00:00 · methodology

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

12 extracted references · 12 canonical work pages

  1. [1]

    A New Wave in Robotics: Survey on Re- cent MmWave Radar Applications in Robotics,

    K. Harlow, H. Jang, T. D. Barfoot, A. Kim, and C. Heckman, “A New Wave in Robotics: Survey on Re- cent MmWave Radar Applications in Robotics,”IEEE Transactions on Robotics, vol. 40, pp. 4544–4560, 2024

    work page 2024

  2. [2]

    Lidar-Level Localiza- tion With Radar? The CFEAR Approach to Accurate, Fast, and Robust Large-Scale Radar Odometry in Di- verse Environments,

    D. Adolfsson, M. Magnusson, A. Alhashimi, A. J. Lilienthal, and H. Andreasson, “Lidar-Level Localiza- tion With Radar? The CFEAR Approach to Accurate, Fast, and Robust Large-Scale Radar Odometry in Di- verse Environments,”IEEE Transactions on Robotics, vol. 39, no. 2, pp. 1476–1495, 2023

    work page 2023

  3. [3]

    ORORA: Outlier-robust radar odometry,

    H. Lim, K. Han, G. Shin, G. Kim, S. Hong, and H. Myung, “ORORA: Outlier-robust radar odometry,” inProceedings of the IEEE International Conference on Robotics and Automation (ICRA), 2023, pp. 2046– 2053

    work page 2023

  4. [4]

    RINO: Accurate, Robust Radar-Inertial Odometry With Non-Iterative Estimation,

    S. Yang, Y . Cao, S. Eben Li, J. Wang, and S. Xu, “RINO: Accurate, Robust Radar-Inertial Odometry With Non-Iterative Estimation,”IEEE Transactions on Automation Science and Engineering, vol. 22, pp. 20 420–20 434, 2025

    work page 2025

  5. [5]

    Boreas: A multi-season autonomous driving dataset,

    K. Burnett et al., “Boreas: A multi-season autonomous driving dataset,”The International Journal of Robotics Research, vol. 42, no. 1-2, pp. 33–42, 2023

    work page 2023

  6. [6]

    1 year, 1000 km: The oxford robotcar dataset,

    W. Maddern, G. Pascoe, C. Linegar, and P. Newman, “1 year, 1000 km: The oxford robotcar dataset,”The International Journal of Robotics Research, vol. 36, no. 1, pp. 3–15, 2017

    work page 2017

  7. [7]

    FoMo: A Multi-Season Dataset for Robot Navigation in For ˆet Montmorency,

    M. Boxan et al., “FoMo: A Multi-Season Dataset for Robot Navigation in For ˆet Montmorency,”arXiv preprint arXiv:2603.08433, 2026

    work page arXiv 2026

  8. [8]

    Pharao: Direct radar odometry using phase correlation,

    Y . S. Park, Y .-S. Shin, and A. Kim, “Pharao: Direct radar odometry using phase correlation,” in2020 IEEE International Conference on Robotics and Automation (ICRA), 2020, pp. 2617–2623

    work page 2020

  9. [9]

    Localization and mapping using only a rotating fmcw radar sensor,

    D. Vivet, P. Checchin, and R. Chapuis, “Localization and mapping using only a rotating fmcw radar sensor,” Sensors, vol. 13, no. 4, pp. 4527–4552, 2013

    work page 2013

  10. [10]

    Radar teach and repeat: Architecture and initial field testing,

    X. Qiao, A. Krawciw, S. Lilge, and T. D. Barfoot, “Radar teach and repeat: Architecture and initial field testing,” in2025 IEEE International Conference on Robotics and Automation (ICRA), IEEE, 2025, pp. 13 021–13 027

    work page 2025

  11. [11]

    Estimation of IMU and MARG ori- entation using a gradient descent algorithm,

    S. O. H. Madgwick, A. J. L. Harrison, and R. Vaidyanathan, “Estimation of IMU and MARG ori- entation using a gradient descent algorithm,” in 2011 IEEE International Conference on Rehabilitation Robotics, 2011, pp. 1–7

    work page 2011

  12. [12]

    Comparing ICP Variants on Real-World Data Sets,

    F. Pomerleau, F. Colas, R. Siegwart, and S. Magnenat, “Comparing ICP Variants on Real-World Data Sets,” Autonomous Robots, vol. 34, no. 3, pp. 133–148, Feb. 2013

    work page 2013