Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 27 of the 27 references it cites were found in a registry or an index; 3 of the 20 sources its claims cite were checked for support.

Probabilistic regression for autonomous terrain relative navigation via multi-modal feature learning

Open access
Kim, Ickbum, Singh, Sandeep · 2024 · Scientific Reports

Abstract

The extension of human spaceflight across an ever-expanding domain, in conjunction with intricate mission architectures demands a paradigm shift in autonomous navigation algorithms, especially for the powered descent phase of planetary landing. Deep learning architectures have previously been explored to perform low-dimensional localization with limited success. Due to the expectations regarding novel algorithms in the context of real missions, the proposed approaches must be rigorously evaluated in extraneous scenarios and demonstrate sufficient robustness. In the current work, a novel formulation is proposed to train CNN-based Deep Learning (DL) models in a multi-layer cascading architecture and utilize the resulting classification probabilities as regression weights to estimate the position of the lander spacecraft. The approach leverages image intensity and depth data provided by multiple sensors on board to accurately determine the spacecraft’s location relative to the observed terrain at a specific altitude. Navigation performance is validated through Monte Carlo analysis, demonstrating the efficacy of the proposed DL architecture and the subsequent state-estimation framework across several simulated scenarios. It shows tremendous promise in extending the multi-modal feature learning approach to realistic missions.

Overview

CoChat Summary

TL;DR

The paper proposes a multi-modal cascading CNN that fuses image intensity and stereo-derived depth to perform terrain relative navigation, then converts classification probabilities into probabilistic position regression. In simulated lunar descent tests, the multi-modal model improves classification accuracy and regression robustness over uni-modal baselines, especially under low illumination and noisy depth.

Key findings

  • At Level 1, test accuracy was 76.4% for intensity-only, 96.8% for depth-only, and 99.3% for intensity + depth.
  • At Levels 2 and 3, the multi-modal CNN maintained accuracy above 98% and above 93%, respectively, while intensity-only accuracy dropped as low as 57.4% in one Level 2 quadrant.
  • In severe tests, low-illumination intensity-only accuracy was 34.7%, noisy-depth-only was 52.3%, and combined low illumination + noisy depth was 65.0%; low illumination supplemented by depth reached 98.3%, and noisy depth with regular intensity reached 65.3%.
  • Regression case at image center (200, 250) gave RMSE of 10.01 pixels (20.02 m) for Centered Prediction and 1.98 pixels (3.96 m) for Mean Neighborhood Prediction; a 2500-run Monte Carlo analysis reported state estimation errors of 14.43 pixels for Centered Prediction and 7.151 pixels for Mean Neighborhood Prediction.

Methods

  • The dataset was generated from the Apollo 16 Digital Terrain Model from the Lunar Reconnaissance Orbiter Camera, assuming nadir-pointed stereo images at 352 m altitude, with image windows of 32, 64, and 128 pixels across three cascading levels.
  • Intensity maps were synthesized using Lambert’s cosine law; training used 10 azimuth/elevation combinations, while test data used 50 equally spaced elevation angles from 0 to 180 degrees, plus an extreme test set with 0–20 degree illumination and truncated Gaussian depth noise.
  • The architecture used a Multi-Modal CNN with separate depth and intensity nodes, softmax classification, temperature scaling, and two probabilistic regression approaches: Centered Prediction and Mean Neighborhood Prediction, evaluated with RMSE and a 2500-run Monte Carlo analysis.

Limitations

  • The paper states that the raw DTM resolution is a bottleneck and that higher-resolution DTMs would be needed to train models for lower altitudes.
  • The paper notes that Mean Neighborhood Prediction is more accurate but computationally expensive: 3.58 s versus 0.266 s for Centered Prediction on an Intel i7-12700 and NVIDIA GeForce RTX 3070 Ti.
  • The paper states that further research using real-life datasets is needed; evident from the text, all results are from simulated datasets, and the combined extreme case of low illumination plus noisy depth still achieved only 65.0% accuracy.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1038/s41598-024-81377-z

Cites

27 of 27 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Fig. 1 Terrain relative navigation overview.
Fig. 1 Terrain relative navigation overview.CC BY · PubMed Central
Fig. 2 Dataset generation example. (Left) Depth map with visualization of image centers and boundaries. (Right) Intensity map.
Fig. 2 Dataset generation example. (Left) Depth map with visualization of image centers and boundaries. (Right) Intensity map.CC BY · PubMed Central
Fig. 3 Image excerpts from the intensity dataset, in the increasing order of Sun elevation angle from left-top to right-bottom.
Fig. 3 Image excerpts from the intensity dataset, in the increasing order of Sun elevation angle from left-top to right-bottom.CC BY · PubMed Central
Fig. 4 Extreme conditions example. (Left) Low illumination intensity. (Right) Noisy depth map.
Fig. 4 Extreme conditions example. (Left) Low illumination intensity. (Right) Noisy depth map.CC BY · PubMed Central
Fig. 5 Architecture of the MMCNN.
Fig. 5 Architecture of the MMCNN.CC BY · PubMed Central
Fig. 6 Joint prediction confidences.
Fig. 6 Joint prediction confidences.CC BY · PubMed Central
Fig. 7 Centered prediction and mean neighborhood prediction comparison.
Fig. 7 Centered prediction and mean neighborhood prediction comparison.CC BY · PubMed Central
Fig. 8 Visualization of Level 2, 3 results. (Top) Level 2. (Bottom) Level 3. (Left) Intensity map. (Middle) Depth map. (Right) Intensity and depth.
Fig. 8 Visualization of Level 2, 3 results. (Top) Level 2. (Bottom) Level 3. (Left) Intensity map. (Middle) Depth map. (Right) Intensity and depth.CC BY · PubMed Central
Fig. 9 Regression Case 1 with image center at (200,250). (Left) Intensity image. (Right) Depth image.
Fig. 9 Regression Case 1 with image center at (200,250). (Left) Intensity image. (Right) Depth image.CC BY · PubMed Central
Fig. 10 Heatmap of regression results. (Left) Centered prediction result with RMSE of 10.01. (Right) Mean neighborhood prediction result with RMSE of 1.98.
Fig. 10 Heatmap of regression results. (Left) Centered prediction result with RMSE of 10.01. (Right) Mean neighborhood prediction result with RMSE of 1.98.CC BY · PubMed Central
Fig. 11 Visualization of 2500-run Monte Carlo Analysis with initial image center at (200, 250) and 3\sigma covariance bounds. (Left) Centered prediction result with mean (191.7, 241.5). (Right) Mean neighborhood prediction with mean (199.9, 246.7).
Fig. 11 Visualization of 2500-run Monte Carlo Analysis with initial image center at (200, 250) and 3\sigma covariance bounds. (Left) Centered prediction result with mean (191.7, 241.5). (Right) Mean neighborhood prediction with mean (199.9, 246.7).CC BY · PubMed Central
Fig. 12 Heatmap of regression results for extreme cases. (Left) Centered prediction result with RMSE of 37.72. (Right) Mean neighborhood prediction result with RMSE of 32.94.
Fig. 12 Heatmap of regression results for extreme cases. (Left) Centered prediction result with RMSE of 37.72. (Right) Mean neighborhood prediction result with RMSE of 32.94.CC BY · PubMed Central
Fig. 13 Regression Case 2 with image center at (100,300). (Left) Intensity image. (Right) Depth image.
Fig. 13 Regression Case 2 with image center at (100,300). (Left) Intensity image. (Right) Depth image.CC BY · PubMed Central
Fig. 14 Heatmap of regression results for near-edge scenario. (Left) Centered prediction result with RMSE of 7.78. (Right) Mean neighborhood prediction result with RMSE of 6.68.
Fig. 14 Heatmap of regression results for near-edge scenario. (Left) Centered prediction result with RMSE of 7.78. (Right) Mean neighborhood prediction result with RMSE of 6.68.CC BY · PubMed Central

Verification

Claims the paper makes

2 Valid · 3 Not valid · 12 Not checked · 50 Own results and interpretations

21 more claims

Create a free account to see all 67 claims and to verify the 2 references still open.

See all claims — free account
  • Probably not a substantial claim. 17
  • It may state two separate claims as one. 29
  • The quote only partly states this claim. 3

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

1 of 1

Citations

Cites

Show all 26 references

Research works printed but not linked: 5

  1. Esty, C. C., Lee, D., Martinez, R. & McCarthy, B. Assessment of cislunar staging orbits to support the artemis iii lunar surface mission. In 45th Annual AAS Guidance, Navigation and Control (GN &C) Conference, AAS 22-762 (2023).
  2. Bussey, B., Clarke, S. W., Jenkins, J. & Bailey, B. E. Nasa’s lunar discovery and explorartion program. In AGU Fall Meeting Abstracts, vol. 2019, PA54B–11 (2019).
  3. Bhaskaran, S. & Hopkins, J. Astrobotic: Peregrine lunar lander technical program update. In 2017 Annual Meeting of the Lunar Exploration Analysis Group, vol. 2041, 5019 (2017).
  4. Sternberg, D. C., Setterfield, T. P., Bailey, E. S., Ansar, A. I. & Johnson, A. E. Demonstration of stereo vision for deorbit descent and landing. vol. 169. Binocular stereo vision system; Flight systems; Landing accuracy; Post processing; Ranging accuracy; Relative navigation; Sensing modalities; Visual odometry, 161–172 (Breckenridge, CO, United states, 2019).
  5. Campbell, T., Furfaro, R., Linares, R. & Gaylor, D. A deep learning approach for optical autonomous planetary relative terrain navigation. Spacefl. Mech. 160, 3293–3302 (2017).

Cited by

See all 6

Metadata

Authors
Kim, I., & Singh, S.
Affiliations
Rensselaer Polytechnic Institute (US)
Venue
Scientific Reports
Published
2024-12-02
Volume
14
Issue
1
Pages
29966-29966
ISSN
2045-2322 (ISSN-L)
OpenAlex
W4404918570
Cited by
6
References
26
Citations by year
2026: 4 · 2025: 2
FWCI
1.03
Type
journal article
Language
en
Open access
Licence
Locations
  • Scientific Reports (publishedVersion, cc-by)
  • Research Square (acceptedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
Rensselaer Polytechnic Institute
Topics
  • Physical Sciences › Engineering › Aerospace Engineering › Spacecraft Dynamics and Control
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Robotics and Sensor-Based Localization
  • Physical Sciences › Computer Science › Artificial Intelligence › Reinforcement Learning in Robotics
Keywords
terrain relative navigation, autonomous navigation, planetary landing, powered descent, probabilistic regression, convolutional neural networks, state estimation, Monte Carlo simulation

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Probabilistic regression for autonomous terrain relative navigation via multi-modal feature learning. https://app.cochat.ai/papers/10.1038/s41598-024-81377-z
The paper
Kim, I., & Singh, S. (2024). Probabilistic regression for autonomous terrain relative navigation via multi-modal feature learning. Scientific Reports. https://doi.org/10.1038/s41598-024-81377-z

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.