pith. sign in

arxiv: 2604.04117 · v1 · submitted 2026-04-05 · 💻 cs.RO · cs.CV· cs.LG

Efficient Onboard Spacecraft Pose Estimation with Event Cameras and Neuromorphic Hardware

Arunkumar Rathinam , Jules Lecomte , Jost Reelsen , Gregor Lenz , Axel von Arnim , Djamila Aouada This is my paper

Pith reviewed 2026-05-13 17:17 UTC · model grok-4.3

classification 💻 cs.RO cs.CVcs.LG
keywords spacecraft pose estimationevent camerasneuromorphic hardware6-DoF estimationlow-power inferencespiking neural networksevent representationsreal-time vision
0
0 comments X

The pith

Neuromorphic hardware runs real-time 6-DoF spacecraft pose estimation from event camera data.

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

The paper shows that compact neural networks can be trained on lightweight event-frame representations to regress spacecraft keypoints and then converted into spiking networks that run directly on neuromorphic processors. Event cameras supply sparse change signals that avoid saturation and blur under the extreme lighting and motion of space, while the hardware exploits those sparse activations for fast, low-energy inference. If the demonstration holds, it supplies a concrete path to onboard perception that meets the latency and power limits of autonomous rendezvous and docking without relying on conventional frame cameras or power-hungry GPUs.

Core claim

Compact MobileNet-style networks trained with quantization-aware methods on event representations from a spacecraft dataset can be mapped to spiking neural networks that execute in real time on neuromorphic hardware, delivering the first end-to-end 6-DoF pose estimates under the low-power and low-latency constraints of spaceflight hardware.

What carries the argument

Quantization-aware training of keypoint regression networks followed by conversion to Akida-compatible spiking networks that process event-frame inputs.

If this is right

  • Real-time inference becomes possible at power levels compatible with small spacecraft buses.
  • Three different event representations can be swapped into the same pipeline with measurable accuracy and latency trade-offs.
  • A heatmap output variant further improves accuracy when mapped to the next hardware generation.
  • The pipeline supplies a template for other vision tasks that must run continuously under tight energy budgets.

Where Pith is reading between the lines

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

  • The same sparse-event plus spiking-network pattern could be tested on terrestrial robots operating in high-contrast or fast-motion scenes.
  • Radiation-tolerant variants of the hardware would allow direct validation of the assumption about flight conditions.
  • Combining the event stream with occasional frame data might recover absolute scale without increasing average power draw.

Load-bearing premise

Models trained and quantized on the chosen dataset will retain usable accuracy when faced with actual flight illumination, motion blur, and radiation effects.

What would settle it

A side-by-side comparison showing that pose error on real orbital imagery exceeds the threshold needed for safe proximity operations while the same model meets the threshold on the training dataset.

Figures

Figures reproduced from arXiv: 2604.04117 by Arunkumar Rathinam, Axel von Arnim, Djamila Aouada, Gregor Lenz, Jost Reelsen, Jules Lecomte.

Figure 1
Figure 1. Figure 1: Cumulative distribution of SPEED scores on the SPADES dataset. (a) shows the impact of quantization on various representations [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Qualitative results on LNES. Rows show samples with [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Comprehensive Quantization and Error Analysis: Top two rows display the performance impact of quantization across different representations. Bottom two rows Mean localization and orientation error binned by ground-truth object distance for full-precision (FP) and quantization-aware trained (QAT) models between V1 and V2 architectures. Shaded regions denote ±1 standard deviation. 7 [PITH_FULL_IMAGE:figures… view at source ↗
Figure 4
Figure 4. Figure 4: Scatter distribution of per-sample localization error versus orientation error for full-precision (FP) and quantization-aware trained [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
read the original abstract

Reliable relative pose estimation is a key enabler for autonomous rendezvous and proximity operations, yet space imagery is notoriously challenging due to extreme illumination, high contrast, and fast target motion. Event cameras provide asynchronous, change-driven measurements that can remain informative when frame-based imagery saturates or blurs, while neuromorphic processors can exploit sparse activations for low-latency, energy-efficient inferences. This paper presents a spacecraft 6-DoF pose-estimation pipeline that couples event-based vision with the BrainChip Akida neuromorphic processor. Using the SPADES dataset, we train compact MobileNet-style keypoint regression networks on lightweight event-frame representations, apply quantization-aware training (8/4-bit), and convert the models to Akida-compatible spiking neural networks. We benchmark three event representations and demonstrate real-time, low-power inference on Akida V1 hardware. We additionally design a heatmap-based model targeting Akida V2 and evaluate it on Akida Cloud, yielding improved pose accuracy. To our knowledge, this is the first end-to-end demonstration of spacecraft pose estimation running on Akida hardware, highlighting a practical route to low-latency, low-power perception for future autonomous space missions.

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 / 1 minor

Summary. The paper claims to present the first end-to-end demonstration of 6-DoF spacecraft pose estimation using event cameras coupled with the BrainChip Akida neuromorphic processor. It trains compact MobileNet-style keypoint regression networks on lightweight event-frame representations from the SPADES dataset, applies 8/4-bit quantization-aware training, converts models to Akida-compatible spiking networks, benchmarks three event representations on Akida V1 hardware for real-time low-power inference, and evaluates a heatmap-based model on Akida V2/Cloud for improved accuracy, positioning this as a practical route to low-latency perception for autonomous space missions.

Significance. If the hardware demonstration and accuracy claims hold under operational conditions, the work would represent a meaningful step toward energy-efficient onboard vision for rendezvous and proximity operations, exploiting event cameras' robustness to extreme illumination and motion while leveraging neuromorphic hardware's sparse activation advantages.

major comments (2)
  1. [Abstract] Abstract and experimental sections: The central claim of real-time, low-power inference on Akida hardware is not supported by any reported quantitative metrics (e.g., pose estimation RMSE in translation/rotation, latency in ms, power in mW, or comparisons against CPU/GPU baselines or non-quantized models), nor by ablation studies on event representations; without these the performance assertions cannot be evaluated.
  2. [Conclusion] Conclusion and evaluation sections: The assertion of a 'practical route' for future missions rests on SPADES simulated events only; no transfer experiments, real event-camera streams, radiation-induced noise tests, or flight-analog illumination/motion conditions are described, leaving the hardware results decoupled from the operational regime invoked in the strongest claim.
minor comments (1)
  1. [Methods] The description of the three event representations and their conversion to Akida SNNs would benefit from explicit pseudocode or a diagram showing the full pipeline from raw events to pose output.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for the constructive feedback on our manuscript. We have revised the experimental and conclusion sections to provide the requested quantitative metrics and to more clearly delineate the simulation-based scope of the current results.

read point-by-point responses

  1. Referee: [Abstract] Abstract and experimental sections: The central claim of real-time, low-power inference on Akida hardware is not supported by any reported quantitative metrics (e.g., pose estimation RMSE in translation/rotation, latency in ms, power in mW, or comparisons against CPU/GPU baselines or non-quantized models), nor by ablation studies on event representations; without these the performance assertions cannot be evaluated.

    Authors: We agree that the original submission lacked explicit numerical reporting of pose RMSE, latency, power draw, and baseline comparisons. The revised manuscript now includes a new experimental subsection with tabulated results for all three event representations on Akida V1, reporting translation/rotation RMSE, inference latency in ms, power consumption in mW, and direct comparisons against CPU and GPU implementations as well as non-quantized floating-point models. Ablation results across the event representations are also provided with these metrics. revision: yes

  2. Referee: [Conclusion] Conclusion and evaluation sections: The assertion of a 'practical route' for future missions rests on SPADES simulated events only; no transfer experiments, real event-camera streams, radiation-induced noise tests, or flight-analog illumination/motion conditions are described, leaving the hardware results decoupled from the operational regime invoked in the strongest claim.

    Authors: We acknowledge that all reported results use the SPADES simulated event dataset and that no real event-camera streams, radiation tests, or flight-analog experiments are included. The revised conclusion now contains an explicit limitations paragraph stating that the hardware demonstration is performed on simulated data, while noting that event cameras' inherent robustness to high dynamic range and motion blur is expected to translate to real space imagery. We have tempered the language around 'practical route' to reflect this scope and added a forward-looking statement on planned real-world validation. revision: partial

standing simulated objections not resolved
  • New experiments involving real event-camera hardware, radiation-induced noise, or flight-analog illumination/motion conditions cannot be added, as they require additional data collection and access beyond the current revision.

Circularity Check

0 steps flagged

No circularity: empirical pipeline with independent experimental content

full rationale

The paper presents an end-to-end empirical demonstration: training MobileNet-style keypoint regressors on the SPADES dataset, applying quantization-aware training, converting to Akida-compatible SNNs, and measuring latency/power on Akida V1/V2 hardware. No mathematical derivation chain exists. No equations are claimed to predict quantities from fitted parameters by construction, no uniqueness theorems are invoked via self-citation, and no ansatz is smuggled through prior work. All load-bearing steps (dataset choice, representation selection, quantization, hardware mapping) are externally verifiable experimental choices whose outcomes are measured rather than defined into existence. The central claim reduces to measured accuracy and power figures on the chosen benchmark, not to a self-referential identity.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 0 invented entities

The central claim rests on neural-network weights fitted to the SPADES dataset, chosen event representations, and hardware-specific quantization and conversion steps; no new physical entities or untested axioms are introduced.

free parameters (2)
  • quantization bit widths
    8-bit and 4-bit precisions selected for Akida compatibility and accuracy-efficiency trade-off.
  • network architecture hyperparameters
    MobileNet-style depth and width choices fitted during training on SPADES.
axioms (1)
  • domain assumption Event-frame representations retain sufficient geometric information for 6-DoF keypoint regression under space illumination conditions.
    Invoked when three event representations are benchmarked without further justification of information loss.

pith-pipeline@v0.9.0 · 5531 in / 1286 out tokens · 38894 ms · 2026-05-13T17:17:33.066992+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

16 extracted references · 16 canonical work pages

  1. [1]

    Nicholson, Andr ´e van Schaik, and Gregory Cohen

    Saeed Afshar, Alison P. Nicholson, Andr ´e van Schaik, and Gregory Cohen. Event-based object detection and tracking for space situational awareness.IEEE Sensors Journal, 20 (24):15117–15132, 2020. 1, 2

    work page 2020

  2. [2]

    Brainchip boosts space heritage by launching akida into low earth orbit

    BrainChip Holdings Ltd. Brainchip boosts space heritage by launching akida into low earth orbit. Press release, 2024. Accessed 2026-02-21. 2, 3

    work page 2024

  3. [3]

    AKD1000 Akida System-on-Chip product brief (version 2.2)

    BrainChip Inc. AKD1000 Akida System-on-Chip product brief (version 2.2). Product brief, 2025. Accessed 2026-02-

    work page 2025

  4. [4]

    Star tracking using an event camera

    Tat-Jun Chin, Sabyasachi Bagchi, Anders Eriksson, Andr ´e van Schaik, and David Suter. Star tracking using an event camera. InProceedings of the IEEE/CVF Confer- 8 ence on Computer Vision and Pattern Recognition Work- shops (CVPRW), 2019. 1, 2

    work page 2019

  5. [5]

    Event-based sensing for space situational aware- ness.The Journal of the Astronautical Sciences, 66:125–141,

    Gregory Cohen, Saeed Afshar, Brittany Morreale, Travis Bessell, Andrew Wabnitz, Mark Rutten, and Andr ´e van Schaik. Event-based sensing for space situational aware- ness.The Journal of the Astronautical Sciences, 66:125–141,

    work page

  6. [6]

    Loihi: A neuromor- phic manycore processor with on-chip learning.IEEE Micro, 38(1):82–99, 2018

    Mike Davies, Narayan Srinivasa, Tsung-Han Lin, Gau- tham Chinya, Yongqiang Cao, Sri Harsha Choday, Geor- gios Dimou, Prasad Joshi, Nabil Imam, Shweta Jain, Yuyun Liao, Chit-Kwan Lin, Andrew Lines, Ruokun Liu, Deepak Mathaikutty, Steve McCoy, Arnab Paul, Jonathan Tse, Guruguhanathan Venkataramanan, Yi-Hsin Weng, Andreas Wild, Yoonseok Yang, and Hong Wang. ...

    work page 2018

  7. [7]

    Furber, Francesco Galluppi, Steve Temple, and Luis A

    Steve B. Furber, Francesco Galluppi, Steve Temple, and Luis A. Plana. The SpiNNaker project.Proceedings of the IEEE, 102(5):652–665, 2014. 2

    work page 2014

  8. [8]

    Davison, J ¨org Conradt, Kostas Daniilidis, and Davide Scaramuzza

    Guillermo Gallego, Tobi Delbr ¨uck, Garrick Orchard, Chiara Bartolozzi, Brian Taba, Andrea Censi, Stefan Leutenegger, Andrew J. Davison, J ¨org Conradt, Kostas Daniilidis, and Davide Scaramuzza. Event-based vision: A survey.IEEE Transactions on Pattern Analysis and Machine Intelligence, 44(1):154–180, 2022. 3

    work page 2022

  9. [9]

    Towards bridging the space domain gap for satellite pose estimation with events.arXiv preprint arXiv:2209.11945,

    Mohsi Jawaid, Ethan Elms, Yasir Latif, and Tat-Jun Chin. Towards bridging the space domain gap for satellite pose estimation with events.arXiv preprint arXiv:2209.11945,

    work page arXiv

  10. [10]

    Low-power ship de- tection in satellite images using neuromorphic hardware

    Gregor Lenz and Douglas McLelland. Low-power ship de- tection in satellite images using neuromorphic hardware. arXiv preprint arXiv:2406.11319, 2024. 3

    work page arXiv 2024

  11. [11]

    SPEED+: Next-generation spacecraft pose estimation dataset for hardware-in-the-loop applications.arXiv preprint arXiv:2110.03101, 2021

    Tae Ha Park, Shashank Sharma, and Simone D’Amico. SPEED+: Next-generation spacecraft pose estimation dataset for hardware-in-the-loop applications.arXiv preprint arXiv:2110.03101, 2021. 1, 2

    work page arXiv 2021

  12. [12]

    T. H. Park, M. M ¨artens, M. Jawaid, Z. Wang, B. Chen, T.- J. Chin, D. Izzo, and S. D’Amico. Satellite pose estima- tion competition 2021: Results and analyses. 204:640–665,

    work page 2021

  13. [13]

    A sur- vey on deep learning-based monocular spacecraft pose esti- mation: Current state, limitations and prospects.Acta Astro- nautica, 212:339–360, 2023

    Leo Pauly, Wassim Rharbaoui, Carl Shneider, Arunkumar Rathinam, Vincent Gaudilli`ere, and Djamila Aouada. A sur- vey on deep learning-based monocular spacecraft pose esti- mation: Current state, limitations and prospects.Acta Astro- nautica, 212:339–360, 2023. 1, 2

    work page 2023

  14. [14]

    Real-time event-based unsupervised feature consolidation and tracking for space situational awareness.Frontiers in Neuroscience, 16:821157, 2022

    Nicholas Ralph, Reuben Joubert, Matthew Jolley, Saeed Af- shar, Gregory Cohen, and Andr ´e van Schaik. Real-time event-based unsupervised feature consolidation and tracking for space situational awareness.Frontiers in Neuroscience, 16:821157, 2022. 1, 2

    work page 2022

  15. [15]

    SPADES: A realistic spacecraft pose estimation dataset using event sensing

    Arunkumar Rathinam, Haytam Qadadri, and Djamila Aouada. SPADES: A realistic spacecraft pose estimation dataset using event sensing. In2024 IEEE International Conference on Robotics and Automation (ICRA), 2024. 1, 2, 3

    work page 2024

  16. [16]

    Reed, Connor Hashemi, Dennis Melamed, Nitesh Menon, Keigo Hirakawa, and Scott McCloskey

    Albert W. Reed, Connor Hashemi, Dennis Melamed, Nitesh Menon, Keigo Hirakawa, and Scott McCloskey. EBS-EKF: Event-based star tracking with an extended kalman filter. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition (CVPR), 2025. 1, 2 9

    work page 2025