CoChat
Sign Up
Cited by 65Spacecraft Dynamics and ControlAerospace Engineering and Control SystemsInertial Sensor and Navigation

Optical navigation for Lunar landing based on Convolutional Neural Network crater detector

Open access
Silvestrini, Stefano, Piccinin, Margherita, Zanotti, Giovanni, Brandonisio, Andrea, Bloise, Ilaria, Feruglio, Lorenzo, Lunghi, Paolo, Lavagna, Michèle +1 more·2022·Aerospace Science and Technology
DOI 10.1016/j.ast.2022.107503
Open this paper in CoChat — free accountOpen PDF

Integrity signals

Registry

Crossref · 10.1016/j.ast.2022.107503

Cites

56 of 57 references verified0 citation errors · 1 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

15 of 31 cited sources checked · 5 supported · 2 flagged · accuracy not yet measured

  • Premisecitation

    State-of-the-art crater detection networks rely heavily on the U-Net architecture to identify landmarks by means of image segmentation.

    “State of the art crater detection networks rely heavily on the so called U-Net architecture [34,35] to identify landmarks by means of image segmentation [20,21,22].”
    Quote in the paperSubstantial claim4 of 5 sources checked
    • cites [34] 34. E. Shelhamer, J. Long, T. Darrell, Fully convolutional net- works for semantic segmentation, IEEE Transactions on Pat- tern Analysis and Machine Intelligence 39 (4) (2017) 640-651. doi:10.1109/TPAMI.2016.2572683. source
    • cites [35] 35. O. Ronneberger, P. Fischer, T. Brox, U-net: Convolutional networks for biomedical image segmentation, in: N. Navab, J. Hornegger, W. M. Wells, A. F. Frangi (Eds.), Medical Image Computing and Computer- Assisted Intervention -MICCAI 2015, Springer International Publish- ing, Cham, 2015, pp. 234-241. source

      The citation to SSD falls in the group attached to identifying landmarks by image segmentation, not the group attached to U-Net. The Model section (p. 3) says SSD detects object classes using bounding boxes and class scores; it does not describe landmark identification by image segmentation. (LLM, conf=0.85)

      Checked by a reasoning model — may be wrong.

      Experimental — accuracy not yet measured
    • cites [20] 20. A. Silburt, M. Ali-Dib, C. Zhu, A. Jackson, D. Valencia, Y. Kissin, D. Tamayo, K. Menou, Lunar crater identification via deep learning, Icarus 317 (2019) 27-38. doi:10.1016/j.icarus.2018.06.022. source

      Section 2.3 (pp. 14–15) describes this crater-detection network as a custom U-Net that predicts a pixel-wise crater-ring mask, supporting its use as an example of segmentation-based detection. The paper does not establish the broader claim that state-of-the-art crater-detection networks generally rely heavily on U-Net; Section 2.4 (p. 16) also says a separate pipeline extracts crater positions and sizes from the predicted masks. (LLM, conf=0.84)

      Experimental — accuracy not yet measured
    • cites [21] 21. L. Downes, T. J. Steiner, J. P. How, Deep Learning Crater Detection for Lunar Terrain Relative Navigation, in: AIAA Scitech 2020 Forum, 2020, pp. 1-12. doi:10.2514/6.2020-1838. source
    • cites [22] 22. L. M. Downes, T. J. Steiner, J. P. How, Lunar terrain relative navi- gation using a convolutional neural network for visual crater detection, in: 2020 American Control Conference (ACC), 2020, pp. 4448-4453. doi:10.23919/ACC45564.2020.9147595. source
  • Premisecitation

    The Larsen measurement extrapolation method does not require knowledge about the measurement matrix and noise distribution matrix until time k.

    “Larsen developed a measurement extrapolation method that does not require knowledge about the two matrices until time k [43].”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [43] 43. T. D. Larsen, N. A. Andersen, O. Ravn, N. K. Poulsen, Incorporation of time delayed measurements in a discrete-time Kalman filter, Proceedings of the IEEE Conference on Decision and Control 4 (December) (1998) 3972-3977. doi:10.1109/cdc.1998.761918. source

      Section 2 (Algorithm) describes Adam’s gradient-based optimization, not Larsen’s measurement extrapolation method. Searches for Larsen, measurement extrapolation, and the claimed timing of knowledge about the two matrices found no passage addressing the claim in our copy. (LLM, conf=0.78)

      Experimental — accuracy not yet measured
  • Premisecitation

    A navigation system can be effective with a minimum frequency of trajectory update of 0.2 Hz.

    “is known that such system can be effective with a minimum frequency of the trajectory update of 0.2 Hz [30,31].”
    Quote in the paperNot a substantial claim2 of 2 sources checked
    • cites [30] 30. P. Lunghi, R. Armellin, P. Di Lizia, M. Lavagna, et al., Semi-analytical adaptive guidance computation based on differential algebra for au- tonomous planetary landing, in: 26th AAS/AIAA Space Flight Me- chanics Meeting, Vol. 158, Univelt, 2016, pp. 2003-2022. source

      Section 5.2.4 (p. 24) says the simulated guidance subsystem recalculates the trajectory every 5 seconds, equivalent to 0.2 Hz. The same section (p. 25) reports good overall performance despite uncertainties. The paper does not establish that 0.2 Hz is the minimum effective trajectory-update frequency. (LLM, conf=0.88)

      Experimental — accuracy not yet measured
    • cites [31] 31. P. Lunghi, M. Lavagna, R. Armellin, A semi-analytical guidance al- gorithm for autonomous landing, Advances in Space Research 55 (11) (2015) 2719-2738. doi:10.1016/j.asr.2015.02.022. source

      Section 5.2.4 (p. 24) says the simulated guidance subsystem recalculates its trajectory every 5 seconds, equivalent to 0.2 Hz. Section 5.2.4 (p. 25) reports good performance in that simulation. The paper does not establish that 0.2 Hz is the minimum effective update frequency. (LLM, conf=0.86)

      Experimental — accuracy not yet measured
  • Findingcitation

    The ODN crater detections are robust to disturbances or modifications of image quality due to Gaussian noise, shot noise, brightness levels, and different illumination conditions.

    “the amount of crater detections that ODN-detector produces is robust to disturbances or modifications of the image quality due to Gaussian noise, shot noise, brightness levels, as well as different illumination conditions.”
    Quote in the paperSubstantial claim2 of 4 sources checked
    • cites [21] 21. L. Downes, T. J. Steiner, J. P. How, Deep Learning Crater Detection for Lunar Terrain Relative Navigation, in: AIAA Scitech 2020 Forum, 2020, pp. 1-12. doi:10.2514/6.2020-1838. source
    • cites [22] 22. L. M. Downes, T. J. Steiner, J. P. How, Lunar terrain relative navi- gation using a convolutional neural network for visual crater detection, in: 2020 American Control Conference (ACC), 2020, pp. 4448-4453. doi:10.23919/ACC45564.2020.9147595. source

      Section VI (page unavailable) reports that LunaNet’s crater detections remain relatively consistent when images are 30% brighter or darker. The paper does not identify its detector as ODN or report tests of Gaussian noise, shot noise, or distinct illumination conditions; Section III mentions robustness to noise only generally. (LLM, conf=0.84)

      Experimental — accuracy not yet measured
    • cites [47] 47. L. M. Downes, T. J. Steiner, J. P. How, Lunar Terrain Rel- ative Navigation Using a Convolutional Neural Network for

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [16] 16. S. Wang, R. Clark, H. Wen, N. Trigoni, Deepvo: Towards end-to- end visual odometry with deep recurrent convolutional neural networks, in: 2017 IEEE International Conference on Robotics and Automation (ICRA), 2017, pp. 2043-2050. doi:10.1109/ICRA.2017.7989236. source

      The Training and Testing section (p. 5) describes DeepVO visual odometry, not ODN crater detection. The VO Results (p. 7) suggest augmenting future training data with noise, but do not report tests of Gaussian noise, shot noise, brightness, or illumination on crater detections. Searches for ODN crater detections and those image-quality conditions found no supporting passage in our copy. (LLM, conf=0.78)

      Experimental — accuracy not yet measured
  • Findingown data

    Preliminary tests performed with the U-Net architecture confirmed a computational burden potentially too high to achieve the target of at least one image processed per second on flight-representative hardware.

    “Preliminary tests performed with such architecture confirmed this trend, with a computational burden potentially too high to achieve the target of at least one image processed per second on flight-representative hardware.”
    Quote in the paperSubstantial claim
  • Classified as probably not a substantial claim. 27
  • May combine two claims. 11
  • The quote only partly states this claim. 1

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The Intel Myriad 2 VPU has already proven effective in flight.

    “that has already proven effective in flight [12,13].”
    Quote in the paperNot a substantial claim2 of 2 sources checked
    • cites [12] 12. G. Giuffrida, L. Diana, F. de Gioia, G. Benelli, G. Meoni, M. Do- nati, L. Fanucci, Cloudscout: A deep neural network for on-board cloud detection on hyperspectral images, Remote Sensing 12 (14) (2020). doi:10.3390/rs12142205. source

      The introduction says the Myriad 2–based algorithm “will be launched,” and Section 2 says the mission aims to demonstrate its usefulness in space. Section 5 adds that the approach still needs validation with HyperScout-2 data. The paper therefore describes planned in-flight demonstration, not effectiveness already proven in flight. (LLM, conf=0.87)

      Checked by a reasoning model — may be wrong.

      Experimental — accuracy not yet measured
    • cites [13] 13. G. Mateo-Garcia, J. Veitch-Michaelis, L. Smith, S. V. Oprea, G. Schu- mann, Y. Gal, A. G. Baydin, D. Backes, Towards global flood mapping onboard low cost satellites with machine learning, Scientific Reports 11 (2021) 7249. doi:10.1038/s41598-021-86650-z. source

      The Introduction says a satellite launched in September 2020 carries the Myriad 2 VPU and that a cloud-detection model was deployed on it. The Results report effective processing on a Myriad 2 chip connected to a Raspberry Pi, but do not establish that the VPU had proven effective in flight. (LLM, conf=0.84)

      Experimental — accuracy not yet measured

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Object detection algorithms using Deep Convolutional Networks have been proposed for UAV obstacle avoidance.

    “Finally, object detection algorithms using Deep Convolutional Networks have been proposed for UAV obstacle avoidance in [26].”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [26] 26. R. Opromolla, G. Fasano, Visual-based obstacle detec- tion and tracking, and conflict detection for small UAS sense and avoid, Aerospace Science and Technology (2021) 107167doi:https://doi.org/10.1016/j.ast.2021.107167. source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    State-of-the-art crater detectors make use of Object Detection Networks and image segmentation.

    “State of the art detectors make use of Object Detection Networks (ODN) [18,19] image segmentation [20,21,22].”
    Quote in the paperNot a substantial claim2 of 5 sources checked
    • cites [18] 18. E. Emami, T. Ahmad, G. Bebis, A. Nefian, T. Fong, Lunar crater de- tection via region-based convolutional neural networks, in: 49th Lunar and Planetary Science Conference, The Woodlands, TX, 2018.
    • cites [19] 19. E. Emami, T. Ahmad, G. Bebis, A. Nefian, T. Fong, Crater detection using unsupervised algorithms and convolutional neural networks, IEEE Transactions on Geoscience and Remote Sensing 57 (8) (2019) 5373- 5383. doi:10.1109/TGRS.2019.2899122. source
    • cites [20] 20. A. Silburt, M. Ali-Dib, C. Zhu, A. Jackson, D. Valencia, Y. Kissin, D. Tamayo, K. Menou, Lunar crater identification via deep learning, Icarus 317 (2019) 27-38. doi:10.1016/j.icarus.2018.06.022. source
    • cites [21] 21. L. Downes, T. J. Steiner, J. P. How, Deep Learning Crater Detection for Lunar Terrain Relative Navigation, in: AIAA Scitech 2020 Forum, 2020, pp. 1-12. doi:10.2514/6.2020-1838. source
    • cites [22] 22. L. M. Downes, T. J. Steiner, J. P. How, Lunar terrain relative navi- gation using a convolutional neural network for visual crater detection, in: 2020 American Control Conference (ACC), 2020, pp. 4448-4453. doi:10.23919/ACC45564.2020.9147595. source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The adopted loss function replicates the original implementation by Liu et al. [36].

    “The adopted loss function replicates the original implementation by Liu et al. [36]”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [36] 36. W. Liu, D. Anguelov, D. Erhan, C. Szegedy, S. Reed, C.-Y. Fu, A. C. Berg, SSD: Single Shot MultiBox Detector, in: B. Leibe, J. Matas, N. Sebe, M. Welling (Eds.), Computer Vision -ECCV 2016, Springer International Publishing, Cham, 2016, pp. 21-37. source

      The paper’s title page (p. 1) identifies Sandler et al., not Liu et al., as its authors. Its ablation study (Section 6.4, p. 8) discusses MobileNetV2 architecture and performance, but the passages found for “Liu,” “loss function,” and “original implementation” do not address a loss function replicated from Liu et al. (LLM, conf=0.75)

      Experimental — accuracy not yet measured

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingreasoning

    The conversion from single channel to RGB comes with null computational cost, because the information of the original channel is replicated 3 times.

    “The conversion from single channel to RGB comes with null computational cost, for the information of the original channel is just replicated 3 times.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingreasoning

    The extrapolation method requires only two matrix multiplications at each time instant and storage of two variables whenever an image is acquired.

    “the extrapolation method always requires only two matrix multiplications at each time instant and the storage of two variables any time an image is acquired.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingnone given

    An inference time on the network of 0.02 s per image on a Nvidia RTX.

    “An inference time on the network of 0.02 s per image on a Nvidia RTX”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingreasoning

    The small peak present in the plot refers to the passage on the North Pole.

    “The small peak present in the plot refers to the passage on the North Pole.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingnone given

    Crater diameters are slightly over-estimated due to ODN post-processing.

    “The crater diameters are slightly over-estimated due to ODN post-processing”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Findingnone given

    The wide range of Sun azimuth is applicable only for high latitudes close to the Poles.

    “Actually, this wide range is applicable only for high latitudes close to the Poles.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingnone given

    The loss function reached its minimum in 150 epochs.

    “the loss function reached its minimum in 150 epochs.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Findingnone given

    No data augmentation was applied or required.

    “No data augmentation was applied or required.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingnone given

    A fine-tuning of the original weights of the last two network branches was sufficient for converging to an optimal solution, while all other weights were maintained fixed at the original value.

    “A fine-tuning of the original weights of the last two network branches was sufficient for converging to an optimal solution, while all the other weights were maintained fixed at the original value.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    Titan X GPU inference time is 57 frames per second with FP32 model precision.

    “Titan X GPU (or 57 frame per seconds, with FP32 model precision) was preliminary assessed.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    The use of a pre-trained MobileNetV2 model allowed transfer learning to speed up the training process, achieving good results even with a relatively small training set.

    “The use of a pre-trained MobileNetV2 model allowed to rely on transfer learning to speed-up the training process, achieving at the same time good results even with a relatively small training set like the one described in Section 3.1.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    On Myriad hardware, with no code optimization, the inference time is 0.4 s per image (or 2.5 frames per second).

    “On Myriad hardware, with no code optimization, the inference time increases to 0.4 s per image (or 2.5 frames per seconds)”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premiseown data

    The U-Net architecture involves a huge number of trainable parameters, with a correspondingly high computational cost.

    “Despite its high accuracy, such architecture involves a huge number of trainable parameters, with a computational cost accordingly high.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Premisenone given

    Transfer learning allows to exploit the efficient set of low level features that pre-trained networks offer.

    “Transfer learning allows to exploit the efficient set of low level features that pre-trained networks offer”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Craters database matching is typically exploited to retrieve the absolute pseudo-measurement.

    “In particular, literature demonstrates that craters database matching is typically exploited to retrieve the absolute pseudo-measurement [10,11].”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [10] 10. L. Singh, S. Lim, On lunar on-orbit vision-based navigation: Ter- rain mapping, feature tracking driven EKF, AIAA Guidance, Nav- igation and Control Conference and Exhibit (August) (2008) 1-18. doi:10.2514/6.2008-6834. source
    • cites [11] 11. C. Xu, X. Huang, M. Li, D. Wang, Landmark database selec- tion for vision-aided inertial navigation in planetary landing missions, Aerospace Science and Technology 118 (2021) 107040. doi:https://doi.org/10.1016/j.ast.2021.107040. URL https://www.sciencedirect.com/science/article/pii/ S1270963821005502 source

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given

    The effect of the Moon's rotation corresponds to a maximum velocity in the Crossrange direction approximately equal to 4.5 m/s during the coasting phase.

    “The effect of the Moon's rotation corresponds to a maximum velocity in the Crossrange direction approximately equal to 4.5 m s -1 during the coasting phase.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Study design and conductreasoning

    By randomly varying environmental variables within the reported ranges, the dataset can cover the wide feature-space expected in the operational scenario.

    “By doing so, the dataset can cover the wide feature-space that is expected in the operational scenario.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Premisecitation

    The lunar impact crater size and distribution has been extracted from [33].

    “The lunar impact crater size and distribution has been extracted from [33].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [33] 33. G. Neukum, B. König, J. Arkani-Hamed, A study of lunar im- pact crater size-distributions, The Moon 12 (2) (1975) 201-229. doi:10.1007/BF00577878. source

  • Probably not a substantial claim: may not pass S4 (based).

    Premisenone given

    The navigation algorithm heavily relies on optical measurements.

    “The navigation algorithm heavily relies on optical measurements.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Study design and conductreasoning

    The synthetic DEM method allows a completely reliable ground-truth for crater position and size, whereas real DEMs contain craters not registered in databases and would not allow a completely reliable training set.

    “This method allows to have a completely reliable ground-truth for the craters position and size, while real DEMs present craters that are not registered in databases and would not allow to build a completely reliable training set.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Premisenone given

    In order to perform end-to-end navigation, craters detected by AI require to be matched with a database.

    “In order to perform end-to-end navigation, craters detected by the AI requires to be matched with a database.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Premisenone given

    During the coasting phase, the spacecraft travels half of the transfer orbit, covering 180 degrees in true anomaly.

    “During the coasting phase, the S/C travels half of the transfer orbit, covering 180 degrees in true anomaly.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    In the South Pole landing scenario, the navigation system can encounter highly varying illumination conditions, with low Sun elevation angle and large shadow areas in the image.

    “In such scenario, the navigation system can encounter highly varying illumination conditions, with low Sun elevation angle in the South Pole region and large shadow areas in the image.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: In Polar Regions, areas at relatively high altitude can experience continuous periods of illumination of several months, whereas some crater bottoms are always in shadow.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Geometrical or feature-based algorithms have been proposed for matching detected craters with a database.

    “Geometrical [23,24] or feature-based [25] algorithms were proposed.”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [23] 23. W. Park, Y. Jung, H. Bang, J. Ahn, Robust crater triangle matching al- gorithm for planetary landing navigation, Journal of Guidance, Control, and Dynamics 42 (2) (2019) 402-410. doi:10.2514/1.G003400. source
    • cites [24] 24. J. qing Chen, P. yuan Cui, H. tao Cui, Automated detection and classifi- cation for craters based on geometric matching, in: J. C. Zarnecki, C. A. Nardell, R. Shu, J. Yang, Y. Zhang (Eds.), International Symposium on Photoelectronic Detection and Imaging 2011: Space Exploration Tech- nologies and Applications, Vol. 8196, International Society for Optics and Photonics, SPIE, 2011, pp. 543 -548. doi:10.1117/12.901020. source
    • cites [25] 25. W. Shao, J. Xie, L. Cao, J. Leng, B. Wang, Crater matching algorithm based on feature descriptor, Advances in Space Research 65 (1) (2020) 616-629. doi:10.1016/j.asr.2019.09.031. source

  • Classified as probably not a substantial claim.

    Premisereasoning

    Since the angle between the Moon rotation axis and the ecliptic is close to 90°, in the Polar Regions the topography plays a crucial role in determining illumination conditions.

    “Since the angle between the Moon rotation axis and the ecliptic is close to 90°, in the Polar Regions the topography plays a crucial role in the determination of the illumination conditions.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Premisenone given

    Increasing the input size affects the inference time, which is potentially crucial for the next implementation of the algorithm into real hardware.

    “increasing the input size affects the inference time, which is potentially crucial for the next implementation of the algorithm into real hardware.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    In prior work, database matching is performed via a nearest neighbor approach, linking the output from the object detector to the navigation filter.

    “In these papers, the database matching is performed via nearest neighbor approach, linking the output from the object detector to the navigation filter.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [10] 10. L. Singh, S. Lim, On lunar on-orbit vision-based navigation: Ter- rain mapping, feature tracking driven EKF, AIAA Guidance, Nav- igation and Control Conference and Exhibit (August) (2008) 1-18. doi:10.2514/6.2008-6834. source
    • cites [11] 11. C. Xu, X. Huang, M. Li, D. Wang, Landmark database selec- tion for vision-aided inertial navigation in planetary landing missions, Aerospace Science and Technology 118 (2021) 107040. doi:https://doi.org/10.1016/j.ast.2021.107040. URL https://www.sciencedirect.com/science/article/pii/ S1270963821005502 source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning

    A possible motivation for the radius overestimation lies in the ODN post-processing stage, in which the crater radius is computed by assuming the bounding box inscribed into a circular shape.

    “A possible motivation lies in the ODN post-processing stage, in which the crater radius is computed by assuming the bounding box inscribed into a circular shape.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: The crater radius is computed by assuming the bounding box is inscribed into a circular shape.

  • May combine two claims.

    Interpretationreasoning

    Transfer learning considerably speeds up the training process, avoiding the burden to learn low-level features inherited from the pre-trained network.

    “Transfer learning considerably speeds-up the training process, avoiding the burden to learn low-level features, inherited from the pre-trained network.”
    Quote in the paperSubstantial claim

  • The quote only partly states this claim.

    Interpretationnone given

    The approximately 200 m estimation error is aligned with the expected performance at these relevant altitudes.

    “which is aligned with the expected performance at these relevant altitudes.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationcitation

    The achieved inference time makes the ODN a promising, and fast, alternative to the iterative, and slower, processes of template matching and thresholding required in classical techniques.

    “the achieved inference time makes the ODN a promising, and fast, alternative to the iterative, and slower, processes of template matching and thresholding required in classical techniques [10].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [10] 10. L. Singh, S. Lim, On lunar on-orbit vision-based navigation: Ter- rain mapping, feature tracking driven EKF, AIAA Guidance, Nav- igation and Control Conference and Exhibit (August) (2008) 1-18. doi:10.2514/6.2008-6834. source

  • Classified as probably not a substantial claim.

    Premisereasoning

    The illumination conditions on ground that the navigation system is expected to encounter are extremely variable, from Sun slightly above the horizon in polar regions to potentially straight illumination with 90° of Sun elevation close to the Lunar Equator.

    “the illumination conditions on ground, especially the inclination of the Sun over the terrain, that the navigation system is expected to encounter are extremely variable, from the Sun slightly above the horizon in polar regions to potentially straight illumination with 90 • of Sun elevation close to the Lunar Equator.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    The radius overestimation does not affect absolute navigation itself, but rather the matching process with the database.

    “Nevertheless, this issue does not affect the absolute navigation itself, but rather the matching process with the database given that the KD-tree search is performed in the three dimensions [x, y, ρ].”
    Quote in the paperNot a substantial claim
37 of 37

  • The paper relies on this without support.

    Traditional vision-based navigation algorithms are highly affected by nonnominal conditions, which include illumination conditions and environmental uncertainties.

    “Traditional vision-based navigation algorithms are highly affected from nonnominal conditions, which comprise illumination conditions and environmental uncertainties.”

  • The paper relies on this without support.

    In Polar Regions, areas at relatively high altitude can experience continuous periods of illumination of several months, whereas some crater bottoms are always in shadow.

    “In fact, areas at relatively high altitude can experience continuous periods of illumination (of several months), whereas some crater bottoms are always in shadow.”

  • The paper relies on this without support.

    The proper behavior of a CNN model when performing inferences on real data is not assured when the training is based on synthetic data.

    “One of the major concerns with CNN and AI in general is their generalization ability, and the proper behavior of the model when performing inferences on real data is not assured when the training is based on synthetic data.”

  • The paper relies on this without support.

    Beyond certain frequencies, the gain in performance due to faster trajectory update becomes irrelevant.

    “Beyond certain frequencies the gain in performance due to the faster update becomes irrelevant.”

  • The authors state this as an assumption.

    The inclination of the initial parking orbit can be tuned to match the latitude of the target landing site, achieving a planar trajectory in the inertial reference frame.

    “in this work is assumed that the inclination of the initial PO can be tuned to match the latitude of the target landing site achieving a planar trajectory (in the inertial reference frame).”

  • The authors state this as an assumption.

    A camera with 40° Field Of View (FOV) and a 1024 × 1024 pixels sensor is the main navigation sensor.

    “A camera with 40° Field Of View (FOV) and a 1024 × 1024 pixels sensor 177 is assumed as main navigation sensor.”

  • The authors state this as an assumption.

    The navigation camera maintains a nominal nadir pointing.

    “the navigation camera is assumed to maintain a nominal nadir pointing”

  • The authors state this as an assumption.

    No particular constraints bound the Sun inclination to a specific range.

    “No particular constraints that bound the Sun inclination to a specific range (like the execution of the whole landing maneuver close to the lunar terminator) are assumed.”

  • The paper relies on this without support.

    Only the Moon gravitational pull is included in the translational dynamics.

    “Since the whole maneuver takes place in close proximity to the lunar ground, only the Moon gravitational pull is included in the translational dynamics.”

  • The authors state this as an assumption.

    The crater radius is computed by assuming the bounding box is inscribed into a circular shape.

    “in which the crater radius is computed by assuming the bounding box inscribed into a circular shape.”

  • The authors state this as an assumption.

    A standard deviation of 1% of the current altitude for altimeter noise reflects the actual behavior of the laser altimeter technology.

    “A standard deviation of 1 % of the current altitude is assumed, reflecting the actual behavior of the laser altimeter technology.”
11 of 11

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.

  • Velocity Jacobian has twice the stated Coriolis coefficient

    Minor

    Equation 4 contains one V ∧ Ω term. Differentiating that term with respect to velocity gives off-diagonal coefficients of ±Ω, but Equation 7 gives ±2Ω.

  • Navigation reference frame is described as both inertial and rotating

    Minor

    The absolute-navigation state is said to be measured in an inertial frame fixed to the Moon. Elsewhere, the paper identifies its lunar-fixed frame as non-inertial because it rotates with the Moon.

2 of 2

Cites

Flagged first, then by citations

Cited by

Sorted by citations

Related

Authors
Silvestrini, S., Piccinin, M., Zanotti, G., Brandonisio, A., Bloise, I., Feruglio, L., Lunghi, P., Lavagna, M., & Varile, M.
Affiliations
Politecnico di Milano (IT)
Corresponding author
Silvestrini, Stefano
ORCID
Venue
Aerospace Science and Technology
Publisher
Elsevier BV
Published
2022-04
Volume
123
Pages
107503
ISSN
1270-9638 (ISSN-L)
DOI
10.1016/j.ast.2022.107503
OpenAlex
W4221084397
Cited by
65
References
52
Citations by year
2026: 7 · 2025: 24 · 2024: 15 · 2023: 14 · 2022: 5
FWCI
6.44
Type
journal article
Language
en
Open access
Licence
Locations
  • Aerospace Science and Technology (publishedVersion)
  • Virtual Community of Pathological Anatomy (University of Castilla La Mancha) (submittedVersion, other-oa)
Funders
European Space Agency
Topics
  • Physical Sciences › Engineering › Aerospace Engineering › Spacecraft Dynamics and Control
  • Physical Sciences › Engineering › Aerospace Engineering › Aerospace Engineering and Control Systems
  • Physical Sciences › Engineering › Aerospace Engineering › Inertial Sensor and Navigation
Keywords
lunar landing, visual navigation, convolutional neural networks, crater detection, object detection, extended Kalman filter, time delay measurement, powered descent, feature extraction

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.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-11.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-11.

Cited sources are checked by 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