Pith. sign in

REVIEW 2 major objections 4 minor 97 references

Robust PnP can be reformulated as an energy minimization that runs on neuromorphic hardware and draws roughly 1% of the power of an embedded CPU at similar runtime and competitive 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 19:47 UTC pith:K35SYI6F

load-bearing objection The central derivation equating inverse iteration to an energy minimization is mathematically false, but the honest hardware demonstration still makes this worth a serious referee. the 2 major comments →

arxiv 2607.16834 v1 pith:K35SYI6F submitted 2026-07-18 cs.CV cs.ET

Robust PnP on a Neuromorphic Processor for Object Pose Estimation

Tam Ngoc-Bang Nguyen , Mohsi Jawaid , Tat-Jun Chin This is my paper
classification cs.CV cs.ET
keywords object pose estimationperspective-n-pointneuromorphic computingspiking neural networkrobust estimationRANSACevent cameraenergy efficiency
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 sets out to prove that perspective-n-point (PnP) — the geometric optimization that recovers an object's pose from 2D-3D correspondences — can be reformulated as a distributed, energy-minimization process that runs on neuromorphic hardware, without sacrificing accuracy. The authors recast the standard direct-linear-transform solver as a constrained quadratic program, implement its primal-dual gradient updates in the neuron/synapse fabric of a neuromorphic chip, and wrap it in a RANSAC-style hypothesize-and-verify loop using gating variables. On hardware, the resulting solver draws a few milliwatts — about one percent of the power of an embedded CPU running a standard robust PnP — while finishing in roughly the same time and matching standard solvers' accuracy in CPU simulation. If the approach holds, the last non-learning component of a two-stage pose-estimation pipeline becomes deployable on low-power edge hardware.

Core claim

The paper's claim is that robust PnP reduces to a constrained energy minimization whose dynamics are exactly the update rules of a neural population, and that this mapping is a concrete deployment: matrix-vector products land in the synaptic weights, states live in neuron potentials, and the fixed coefficients of the DLT data matrix are frozen at runtime. Random hypothesis generation is achieved by binary gating variables that switch off correspondences, and an inlier-counting verification layer scores each hypothesis. The authors report that on a neuromorphic processor, the solver draws roughly 98–99% less dynamic power than an embedded CPU running a standard robust PnP at comparable runtim

What carries the argument

The mechanism is a reinterpretation of inverse iteration for the direct linear transform as a constrained quadratic program, solved by alternate updates of a 12-dimensional state θ (the vectorized camera matrix) and a Lagrange multiplier λ enforcing the norm constraint. The update equations are mapped onto a neuromorphic chip's synapse-neuron structure: matrix-vector products become synapse weights, the running state is the neuron potential, and the fixed coefficients in the data matrix are frozen in the synaptic weights at runtime. Randomized hypothesis generation is implemented by gating variables that selectively shut off correspondences, and a verification module computes algebraic resid

Load-bearing premise

The load-bearing premise is that the fixed-step-size primal-dual gradient updates converge to the constrained minimum of the DLT objective within the allotted inner iterations for every input; the paper gives no proof of this convergence, and its own hardware results show that when numerical precision drops, the approximation errors become noticeable.

What would settle it

Construct a synthetic PnP instance with known ground truth and run the exact-arithmetic dynamics with the paper's fixed step sizes; if the iterates do not converge to the DLT right-singular-vector solution within the allotted outer iterations, or if a run on neuromorphic hardware diverges from the CPU simulation on the same input, the central claim fails.

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

If this is right

  • Two-stage object-pose estimation can in principle run entirely on neuromorphic hardware: the landmark-regressor SNN and the robust PnP solver no longer need a CPU.
  • On the measured synthetic instances, dynamic power draw on the neuromorphic chip was about 98–99% lower than the embedded CPU reference at similar runtime, implying roughly 100× higher energy efficiency for the PnP step.
  • NeuroPnP's accuracy in CPU simulation tracks standard robust PnP baselines across outlier ratios from 10% to 50% on two event-camera datasets.
  • The neuron-count complexity is linear in the number of correspondences (16N+15), so the architecture does not explode combinatorially with input size.

Where Pith is reading between the lines

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

  • The same energy-minimization mapping could generalize to other geometric estimation tasks with manifold constraints (e.g., rotation averaging or essential-matrix estimation), potentially carrying over the ~100× efficiency gain, since the underlying dynamics are generic quadratic energy descent.
  • A testable extension is replacing the fixed step sizes with adaptive or normalized updates, which might restore the convergence lost under low-precision arithmetic and make hardware accuracy match CPU simulation without new hardware.
  • The gating mechanism performs subset sampling in place, suggesting the same technique could implement other combinatorial hypothesis generators (e.g., guided or progressive sampling) directly in spiking dynamics.

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

2 major / 4 minor

Summary. The paper proposes NeuroPnP, a robust PnP solver intended for deployment on Intel Loihi 2. It casts DLT inverse iteration as a constrained quadratic energy minimization, implements the minimization with primal-dual neuronal dynamics, and embeds it in a hypothesize-and-verify loop with a gating mechanism for outlier rejection. A companion SNN, SPose, is introduced for 2D landmark regression from event frames, and the two components are combined into a two-stage neuromorphic OPE pipeline. Experiments are conducted in Lava CPU simulation on E-POSE and FRESH data, and on Loihi 2 hardware using 20 synthetic instances with N=15 integer correspondences. The paper reports competitive accuracy in simulation and roughly 1% of an embedded CPU's power draw on Loihi 2, while acknowledging numerical precision limitations.

Significance. If the mathematical formulation were sound, NeuroPnP would be a valuable first demonstration of a fully neuromorphic two-stage OPE pipeline, and the paper's explicit treatment of hardware constraints is commendable. The authors also follow good empirical practice by fixing most hyperparameters across experiments and comparing against standard OpenCV baselines on public datasets. However, the central equivalence between inverse iteration DLT and the proposed quadratic energy in §IV-A is algebraically incorrect, and the convergence of the proposed dynamics is not established. These issues block acceptance in the current form. The appended limitations in §VII-A are honest, but they do not repair the load-bearing derivation error.

major comments (2)
  1. [§IV-A, Eq. (11)] Even if Eq. (9) were accepted, the paper asserts without proof or reference that the projected primal-dual updates (11) converge to the constrained minimizer within G=50 or 1000 steps at fixed α=0.005 and β=0.1. The constraint ||θ||=M is nonconvex, and the algorithm is a gradient descent-ascent scheme for a saddle-point problem; convergence to the global constrained minimum is not automatic. No step-size condition or Lyapunov argument is given. This is load-bearing because the pose-generation layer's correctness rests on this convergence.
  2. [§VI-C and §VII-A] The Loihi 2 evaluation is confined to 20 synthetic instances with N=15 integer correspondences and 3–6 outliers, and the reported AUC is lower than the CPU baseline. The authors admit in §IV-E and §VII-A that fixed-point arithmetic introduces noticeable approximation errors and that larger practical instances cannot be handled. Consequently, the headline claim of ~1% power draw with 'competitive accuracy' is demonstrated only on a toy regime, not on the realistic E-POSE/FRESH problems used in the CPU experiments. The energy-efficiency conclusion therefore needs to be qualified, or the hardware evaluation needs to be scaled up.
minor comments (4)
  1. [Abstract] Typo: 'neuromophic' should be 'neuromorphic'; the sentence 'indicate the higher energy efficiency our neuromorphic robust PnP' is missing 'of'.
  2. [Figs. 6–7] The horizontal axis is labeled 'Outlier Ratio (%)' but the tick values are 0.1–0.5; the text (§VI-B) uses η=10–50%. Please make the units consistent.
  3. [§IV-A, Eqs. (10)–(16)] The neuronal-dynamics notation in (12)–(16) is terse; in particular, the dimensions and signs in Σ_d and σ are not fully explained. A short derivation of how (11) becomes (12)–(13) would improve readability.
  4. [§V-C] 'up to ≈12.6 Million spiking neurons' appears to refer to parameters/units in a simulated SNN rather than physical neurons; please clarify.

Circularity Check

0 steps flagged

No significant circularity: NeuroPnP's derivation and evaluation are self-contained, with only minor non-load-bearing self-citations.

full rationale

The paper's central claim is a neuromorphic-deployable robust PnP algorithm based on inverse iteration [89], an external method, and an energy-minimization reformulation. The hyperparameters (α, β, M, K, G, L) are fixed before experiments and are not fitted to the target poses; the inlier threshold ε_in is tuned per input but is shared identically with all competitor methods, so it does not selectively favor NeuroPnP. Accuracy is evaluated against external baselines (OpenCV RANSAC + DLT/LM and + EPnP) on public datasets E-POSE and FRESH, with NeuroPnP simulated on CPU and compared against an embedded-CPU implementation on the Loihi 2 hardware. The paper's own self-citations (e.g., [57], [68], [75], [80], [92]) appear in related-work context, as dataset or metric sources, or as examples of neuromorphic energy minimization; none is load-bearing for the correctness of the proposed algorithm. The 'primary insight' in Eqs. (8)–(9), while mathematically questionable—the stationarity conditions of the two constrained objectives differ unless (Q−µI) is a scalar or projection-like operator—is a correctness/derivation error, not a circularity: it does not make the output equivalent to the input by construction. The paper also honestly acknowledges hardware precision limitations and the gap between simulation and Loihi 2 results. The derivation chain does not reduce to fitted parameters or self-citation, so there is no significant circularity; the score reflects only the presence of minor self-citations in background and benchmarking materials.

Axiom & Free-Parameter Ledger

8 free parameters · 7 axioms · 0 invented entities

No new physical entities are introduced. The free parameters are optimizer hyperparameters and preprocessing constants; none are fitted to ground-truth poses. The main unstated premise is the convergence of the energy-minimization dynamics, which is neither proven nor formally analyzed. The fixed-point hardware assumption is also load-bearing but explicitly acknowledged.

free parameters (8)
  • α (step size for θ update) = 0.005
    Chosen by hand; affects convergence speed and stability of the energy-minimization dynamics in Eq. (11).
  • β (dual update rate for λ) = 0.1
    Chosen by hand; controls how strongly the norm constraint is enforced.
  • M (norm of p) = 1 (CPU), 10 (Loihi 2)
    Scale of the DLT solution; must be set to avoid trivial solution and to fit fixed-point range on Loihi.
  • K (number of inverse iterations) = 50 (CPU), 10 (Loihi 2)
    Number of outer iterations for DLT via inverse iteration; larger values improve convergence to the least-significant singular vector.
  • G (update steps per inverse iteration) = 1000 (CPU), 50 (Loihi 2)
    Number of gradient updates for the constrained least-squares subproblem; smaller on hardware due to time constraints.
  • L (hypothesize-and-verify loops) = 100
    RANSAC iterations; fixed for all methods.
  • α_exp, β_exp (fixed-point scaling exponents) = 10 and 8
    Used to convert floating-point α and β to integers on Loihi 2, influencing numerical precision.
  • ε_in (inlier threshold) = tuned per input
    Inlier/outlier threshold; tuned per input but shared with all competitor methods, so it is not a NeuroPnP-specific advantage.
axioms (7)
  • standard math Direct Linear Transformation (DLT) is a valid formulation of PnP; the least significant singular vector of A gives the camera matrix (up to scale).
    Invoked in §III-A(b); standard in computer vision.
  • standard math Inverse iteration with shift μ close to 0 converges to the eigenvector of Q=A^T A with eigenvalue closest to μ, i.e., the least significant singular vector of A.
    Used in §IV-A; standard inverse iteration [89], but only if the updates actually reach the true eigenvector.
  • ad hoc to paper The primal-dual gradient dynamics in Eq. (11) converge to the constrained minimum of Eq. (8) within G steps for all inputs.
    No convergence proof is provided; the paper states 'Once equilibrium is attained, the state of θ is installed as p_{k+1}' (§IV-A). This is load-bearing and unproven.
  • domain assumption Sampling each correspondence independently with probability 6/N yields sufficiently many all-inlier minimal subsets within L loops.
    In §IV-B, the paper argues that deviations (more or fewer than 6 points) merely cause noise or lower chance, and more iterations mitigate this. Standard RANSAC assumption but weaker than exact minimal sampling.
  • ad hoc to paper Fixed-point arithmetic on Loihi 2 (8-bit synapses, integer-only operations) preserves enough precision for the energy-minimization dynamics to converge to useful pose estimates.
    The authors acknowledge 'noticeable approximation errors are inevitable' (§IV-E) and restrict hardware tests to small integer instances; the assumption is load-bearing for the hardware claim.
  • domain assumption Accumulating event data into frames of 50 ms windows (5×10 ms chunks) preserves sufficient information for 2D landmark regression.
    Introduced in §V-A; a preprocessing choice that may discard temporal event information.
  • domain assumption SimCC coordinate classification is a suitable representation for 2D landmark regression from event frames.
    Used in §V-B; inherited from [91] and adapted with a spiking backbone.

pith-pipeline@v1.3.0-alltime-deepseek · 19043 in / 10401 out tokens · 97948 ms · 2026-08-01T19:47:15.093800+00:00 · methodology

0 comments
read the original abstract

Neuromorphic computing is gaining attention in robotic perception due to its higher energy efficiency. While neural network-based methods can more readily exploit the distributed and parallelized structure of neuromorphic computers, crafting neuromorphic solutions for non-learning tasks is less straightforward. This hampers the usage of neuromorphic computing for perception pipelines that depend on both learning and non-learning components, such as object pose estimation (OPE) where state-of-the-art methods use a deep network to predict 2D landmarks and nonlinear optimization to solve perspective-n-point (PnP). In this paper, we propose a novel neuromorphic-deployable formulation for robust PnP, where given outlier-prone 2D-3D correspondences, the object pose with the largest number of inliers is determined. Underpinning our method is a distributed algorithm for robust least squares estimation of rigid body pose that can be executed on a neuromorphic processor. We also design a spiking neural network (SNN) to predict 2D landmarks from event data, where the main layers of the SNN were designed according to the principles of spiking neurons. Overall, our work enables neuromorphic treatment of the major stages of an OPE pipeline, from event sensing and learned landmark prediction, to geometric optimization for robust PnP. Results on neuromophic hardware (Intel Loihi 2) indicate the higher energy efficiency our neuromorphic robust PnP, while achieving competitive accuracy.

Figures

Figures reproduced from arXiv: 2607.16834 by Mohsi Jawaid, Tam Ngoc-Bang Nguyen, Tat-Jun Chin.

Figure 1
Figure 1. Figure 1: Contrasting end-to-end (left) and two-stage (right) neuromorphic OPE pipelines. An end-to-end method trains an SNN [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Basic schematic of a neuromorphic core in Intel Loihi [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Architecture of the proposed NeuroPnP algorithm. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: SPose architecture. Each grid represents a binary [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Performance comparison between NeuroPnP and two robust PnP baselines across different outlier ratios on E-POSE. [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Performance comparison between NeuroPnP and two robust PnP baselines across different outlier ratios on FRESH. [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Comparing accuracy, power draw and runtime of [PITH_FULL_IMAGE:figures/full_fig_p008_8.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

97 extracted references · 10 linked inside Pith

  1. [1]

    The moped framework: Object recognition and pose estimation for manipulation,

    A. Collet, M. Martinez, and S. S. Srinivasa, “The moped framework: Object recognition and pose estimation for manipulation,”IJRR, vol. 30, no. 10, pp. 1284–1306, 2011

    2011

  2. [2]

    Are we ready for autonomous driving? the kitti vision benchmark suite,

    A. Geiger, P. Lenz, and R. Urtasun, “Are we ready for autonomous driving? the kitti vision benchmark suite,” in2012 IEEE conference on computer vision and pattern recognition. IEEE, 2012, pp. 3354–3361

    2012

  3. [3]

    Multi-view 3d object detection network for autonomous driving,

    X. Chen, H. Ma, J. Wan, B. Li, and T. Xia, “Multi-view 3d object detection network for autonomous driving,” inProceedings of the IEEE conference on Computer Vision and Pattern Recognition, 2017, pp. 1907–1915

    2017

  4. [4]

    Pointfusion: Deep sensor fusion for 3d bounding box estimation,

    D. Xu, D. Anguelov, and A. Jain, “Pointfusion: Deep sensor fusion for 3d bounding box estimation,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 244–253

    2018

  5. [5]

    Autonomous rendezvous, capture and in-space assembly: past, present and future,

    D. Zimpfer, P. Kachmar, and S. Tuohy, “Autonomous rendezvous, capture and in-space assembly: past, present and future,” in1st Space exploration conference: continuing the voyage of discovery, 2005, p. 2523

    2005

  6. [6]

    A review of cooperative and uncooperative spacecraft pose determination techniques for close-proximity operations,

    R. Opromolla, G. Fasano, G. Rufino, and M. Grassi, “A review of cooperative and uncooperative spacecraft pose determination techniques for close-proximity operations,”Progress in Aerospace Sciences, vol. 93, pp. 53–72, 2017

    2017

  7. [7]

    Bb8: A scalable, accurate, robust to partial occlusion method for predicting the 3d poses of challenging objects without using depth,

    M. Rad and V . Lepetit, “Bb8: A scalable, accurate, robust to partial occlusion method for predicting the 3d poses of challenging objects without using depth,” inProceedings of the IEEE international confer- ence on computer vision, 2017, pp. 3828–3836

    2017

  8. [8]

    Cullnet: Calibrated and pose aware confidence scores for object pose estimation,

    K. Gupta, L. Petersson, and R. Hartley, “Cullnet: Calibrated and pose aware confidence scores for object pose estimation,” inProceedings of the IEEE/CVF international conference on computer vision workshops, 2019, pp. 0–0

    2019

  9. [9]

    Hybridpose: 6d object pose estimation under hybrid representations,

    C. Song, J. Song, and Q. Huang, “Hybridpose: 6d object pose estimation under hybrid representations,” inProceedings of the IEEE/CVF confer- ence on computer vision and pattern recognition, 2020, pp. 431–440

    2020

  10. [10]

    Zebrapose: Coarse to fine surface encoding for 6dof object pose estimation,

    Y . Su, M. Saleh, T. Fetzer, J. Rambach, N. Navab, B. Busam, D. Stricker, and F. Tombari, “Zebrapose: Coarse to fine surface encoding for 6dof object pose estimation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 6738–6748

    2022

  11. [11]

    Zs6d: Zero-shot 6d object pose estimation using vision transformers,

    P. Ausserlechner, D. Haberger, S. Thalhammer, J.-B. Weibel, and M. Vincze, “Zs6d: Zero-shot 6d object pose estimation using vision transformers,” in2024 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2024, pp. 463–469

    2024

  12. [12]

    Rnnpose: 6-dof object pose estimation via recurrent correspondence field estimation and pose optimization,

    Y . Xu, K.-Y . Lin, G. Zhang, X. Wang, and H. Li, “Rnnpose: 6-dof object pose estimation via recurrent correspondence field estimation and pose optimization,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 46, no. 7, pp. 4669–4683, 2024

    2024

  13. [13]

    A review of solutions for perspective-n-point problem in camera pose estimation,

    X. X. Lu, “A review of solutions for perspective-n-point problem in camera pose estimation,” inJournal of Physics: Conference Series, vol. 1087, no. 5. IOP Publishing, 2018, p. 052009

    2018

  14. [14]

    A survey on perspective-n-point problem,

    S. Pan and X. Wang, “A survey on perspective-n-point problem,” in2021 40th Chinese control conference (CCC). IEEE, 2021, pp. 2396–2401

    2021

  15. [15]

    Picopose: Progressive pixel-to- pixel correspondence learning for novel object pose estimation,

    L. Liu, J. Lin, Z. Liu, and K. Jia, “Picopose: Progressive pixel-to- pixel correspondence learning for novel object pose estimation,”arXiv preprint arXiv:2504.02617, 2025

    Pith/arXiv arXiv 2025

  16. [16]

    Hccepose (bf): Predicting front & back surfaces to construct ultra-dense 2d-3d correspondences for pose estimation,

    Y . Wang, M. Hu, H. Li, and C. Luo, “Hccepose (bf): Predicting front & back surfaces to construct ultra-dense 2d-3d correspondences for pose estimation,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2025, pp. 7166–7175

    2025

  17. [17]

    Opportunities for neuromorphic computing algorithms and applications,

    C. D. Schuman, S. R. Kulkarni, M. Parsa, J. P. Mitchell, P. Date, and B. Kay, “Opportunities for neuromorphic computing algorithms and applications,”Nature Computational Science, vol. 2, no. 1, pp. 10–19, 2022

    2022

  18. [18]

    Neuromorphic silicon neurons and large- scale neural networks: challenges and opportunities,

    C.-S. Poon and K. Zhou, “Neuromorphic silicon neurons and large- scale neural networks: challenges and opportunities,”Frontiers in neu- roscience, vol. 5, p. 108, 2011

    2011

  19. [19]

    A million spiking-neuron integrated circuit with a scalable communication network and interface,

    P. A. Merolla, J. V . Arthur, R. Alvarez-Icaza, A. S. Cassidy, J. Sawada, F. Akopyan, B. L. Jackson, N. Imam, C. Guo, Y . Nakamuraet al., “A million spiking-neuron integrated circuit with a scalable communication network and interface,”Science, vol. 345, no. 6197, pp. 668–673, 2014

    2014

  20. [20]

    Efficient neuromorphic signal processing with loihi 2,

    G. Orchard, E. P. Frady, D. B. D. Rubin, S. Sanborn, S. B. Shrestha, F. T. Sommer, and M. Davies, “Efficient neuromorphic signal processing with loihi 2,” in2021 IEEE Workshop on Signal Processing Systems (SiPS). IEEE, 2021, pp. 254–259

    2021

  21. [21]

    Loihi: A neuromorphic manycore processor with on-chip learning,

    M. Davies, N. Srinivasa, T.-H. Lin, G. Chinya, Y . Cao, S. H. Choday, G. Dimou, P. Joshi, N. Imam, S. Jainet al., “Loihi: A neuromorphic manycore processor with on-chip learning,”Ieee Micro, vol. 38, no. 1, pp. 82–99, 2018

    2018

  22. [22]

    Advancing neuromorphic com- puting with loihi: A survey of results and outlook,

    M. Davies, A. Wild, G. Orchard, Y . Sandamirskaya, G. A. F. Guerra, P. Joshi, P. Plank, and S. R. Risbud, “Advancing neuromorphic com- puting with loihi: A survey of results and outlook,”Proceedings of the IEEE, vol. 109, no. 5, pp. 911–934, 2021

    2021

  23. [23]

    Networks of spiking neurons: the third generation of neural network models,

    W. Maass, “Networks of spiking neurons: the third generation of neural network models,”Neural networks, vol. 10, no. 9, pp. 1659–1671, 1997

    1997

  24. [24]

    Training spiking neural JOURNAL OF LATEX CLASS FILES, VOL. 18, NO. 9, SEPTEMBER 2020 10 networks using lessons from deep learning,

    J. K. Eshraghian, M. Ward, E. O. Neftci, X. Wang, G. Lenz, G. Dwivedi, M. Bennamoun, D. S. Jeong, and W. D. Lu, “Training spiking neural JOURNAL OF LATEX CLASS FILES, VOL. 18, NO. 9, SEPTEMBER 2020 10 networks using lessons from deep learning,”Proceedings of the IEEE, vol. 111, no. 9, pp. 1016–1054, 2023

    2020

  25. [25]

    A review of non-cognitive applications for neuromorphic computing,

    J. B. Aimone, P. Date, G. A. Fonseca-Guerra, K. E. Hamilton, K. Henke, B. Kay, G. T. Kenyon, S. R. Kulkarni, S. M. Mniszewski, M. Parsaet al., “A review of non-cognitive applications for neuromorphic computing,” Neuromorphic Computing and Engineering, vol. 2, no. 3, p. 032003, 2022

    2022

  26. [26]

    Detection of fast-moving objects with neuromorphic hardware,

    A. Ziegler, K. Vetter, T. Gossard, J. Tebbe, S. Otte, and A. Zell, “Detection of fast-moving objects with neuromorphic hardware,” in2025 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2025, pp. 8709–8717

    2025

  27. [27]

    Neuromorphic quadratic programming for efficient and scal- able model predictive control: Towards advancing speed and energy efficiency in robotic control,

    A. R. Mangalore, G. A. F. Guerra, S. R. Risbud, P. Stratmann, and A. Wild, “Neuromorphic quadratic programming for efficient and scal- able model predictive control: Towards advancing speed and energy efficiency in robotic control,”IEEE Robotics & Automation Magazine, vol. 32, no. 2, pp. 69–79, 2024

    2024

  28. [28]

    A compact neuromorphic system for ultra–energy-efficient, on-device robot localization,

    A. D. Hines, M. Milford, and T. Fischer, “A compact neuromorphic system for ultra–energy-efficient, on-device robot localization,”Science Robotics, vol. 10, no. 103, p. eads3968, 2025

    2025

  29. [29]

    Visual odometry with neuromorphic resonator networks,

    A. Renner, L. Supic, A. Danielescu, G. Indiveri, E. P. Frady, F. T. Sommer, and Y . Sandamirskaya, “Visual odometry with neuromorphic resonator networks,”Nature Machine Intelligence, vol. 6, no. 6, pp. 653– 663, 2024

    2024

  30. [30]

    Spiking neural network on neuromorphic hardware for energy-efficient unidimensional slam,

    G. Tang, A. Shah, and K. P. Michmizos, “Spiking neural network on neuromorphic hardware for energy-efficient unidimensional slam,” in2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2019, pp. 4176–4181

    2019

  31. [31]

    Event-driven vision and control for uavs on a neuromorphic chip,

    A. Vitale, A. Renner, C. Nauer, D. Scaramuzza, and Y . Sandamirskaya, “Event-driven vision and control for uavs on a neuromorphic chip,” in2021 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2021, pp. 103–109

    2021

  32. [32]

    Fully autonomous neuromorphic navigation and dynamic obstacle avoidance,

    X. Shang, L. Pengwei, X. Wang, J. Zhao, H. Ge, B. Dong, and X. Yang, “Fully autonomous neuromorphic navigation and dynamic obstacle avoidance,” inThe Thirty-ninth Annual Conference on Neural Information Processing Systems, 2025

    2025

  33. [33]

    Fully neuromorphic vision and control for autonomous drone flight,

    F. Paredes-Vall ´es, J. J. Hagenaars, J. Dupeyroux, S. Stroobants, Y . Xu, and G. C. de Croon, “Fully neuromorphic vision and control for autonomous drone flight,”Science Robotics, vol. 9, no. 90, p. eadi0591, 2024

    2024

  34. [34]

    Neuromorphic attitude estimation and control,

    S. Stroobants, C. De Wagter, and G. C. De Croon, “Neuromorphic attitude estimation and control,”IEEE Robotics and Automation Letters, 2025

    2025

  35. [35]

    Spiking monocular event based 6d pose estimation for space application,

    J. Courtois, B. Miramond, and A. Pegatoquet, “Spiking monocular event based 6d pose estimation for space application,” inSPAICE2024: The First Joint European Space Agency / IAA Conference on AI in and for Space, 2024, pp. 394–399

    2024

  36. [36]

    Efficient onboard spacecraft pose estimation with event cameras and neuromorphic hardware,

    A. Rathinam, J. Lecomte, J. Reelsen, G. Lenz, A. von Arnim, and D. Aouada, “Efficient onboard spacecraft pose estimation with event cameras and neuromorphic hardware,”arXiv preprint arXiv:2604.04117, 2026

    Pith/arXiv arXiv 2026

  37. [37]

    Towards bridging the space domain gap for satellite pose estimation using event sensing,

    M. Jawaid, E. Elms, Y . Latif, and T.-J. Chin, “Towards bridging the space domain gap for satellite pose estimation using event sensing,” in2023 IEEE International Conference on Robotics and Automation (ICRA), 2023

    2023

  38. [38]

    Spikingjelly: An open-source machine learning infrastructure platform for spike-based intelligence,

    W. Fang, Y . Chen, J. Ding, Z. Yu, T. Masquelier, D. Chen, L. Huang, H. Zhou, G. Li, and Y . Tian, “Spikingjelly: An open-source machine learning infrastructure platform for spike-based intelligence,”Science Advances, vol. 9, no. 40, p. eadi1480, 2023

    2023

  39. [39]

    Deep residual learning in spiking neural networks,

    W. Fang, Z. Yu, Y . Chen, T. Huang, T. Masquelier, and Y . Tian, “Deep residual learning in spiking neural networks,”Advances in Neural Information Processing Systems, vol. 34, pp. 21 056–21 069, 2021

    2021

  40. [40]

    Advancing spiking neural networks toward deep residual learning,

    Y . Hu, L. Deng, Y . Wu, M. Yao, and G. Li, “Advancing spiking neural networks toward deep residual learning,”IEEE transactions on neural networks and learning systems, vol. 36, no. 2, pp. 2353–2367, 2024

    2024

  41. [41]

    A spiking neural network for image segmentation,

    K. Patel, E. Hunsberger, S. Batir, and C. Eliasmith, “A spiking neural network for image segmentation,”arXiv preprint arXiv:2106.08921, 2021

    Pith/arXiv arXiv 2021

  42. [42]

    Spike2former: Efficient spiking transformer for high-performance image segmentation,

    Z. Lei, M. Yao, J. Hu, X. Luo, Y . Lu, B. Xu, and G. Li, “Spike2former: Efficient spiking transformer for high-performance image segmentation,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 39, no. 2, 2025, pp. 1364–1372

    2025

  43. [43]

    Spiking-yolo: spiking neural network for energy-efficient object detection,

    S. Kim, S. Park, B. Na, and S. Yoon, “Spiking-yolo: spiking neural network for energy-efficient object detection,” inProceedings of the AAAI conference on artificial intelligence, vol. 34, no. 07, 2020, pp. 11 270–11 277

    2020

  44. [44]

    Object detection with spiking neural networks on automotive event data,

    L. Cordone, B. Miramond, and P. Thierion, “Object detection with spiking neural networks on automotive event data,” in2022 International Joint Conference on Neural Networks (IJCNN). IEEE, 2022, pp. 1–8

    2022

  45. [45]

    Deep directly- trained spiking neural networks for object detection,

    Q. Su, Y . Chou, Y . Hu, J. Li, S. Mei, Z. Zhang, and G. Li, “Deep directly- trained spiking neural networks for object detection,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 6555–6565

    2023

  46. [46]

    Integer-valued training and spike-driven inference spiking neural network for high-performance and energy-efficient object detection,

    X. Luo, M. Yao, Y . Chou, B. Xu, and G. Li, “Integer-valued training and spike-driven inference spiking neural network for high-performance and energy-efficient object detection,” inEuropean Conference on Computer Vision. Springer, 2024, pp. 253–272

    2024

  47. [47]

    Brain-inspired spiking neural networks for energy-efficient object detection,

    Z. Li, T. Gao, Y . An, T. Chen, J. Zhang, Y . Wen, M. Liu, and Q. Zhang, “Brain-inspired spiking neural networks for energy-efficient object detection,” inProceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 3552–3562

    2025

  48. [48]

    Efficient event-based object detection: A hybrid neural network with spatial and temporal attention,

    S. H. Ahmed, J. Finkbeiner, and E. Neftci, “Efficient event-based object detection: A hybrid neural network with spatial and temporal attention,” inProceedings of the Computer Vision and Pattern Recognition Confer- ence, 2025, pp. 13 970–13 979

    2025

  49. [49]

    Neuromorphic optical flow and real-time implementation with event cameras,

    Y . Schnider, S. Wo ´zniak, M. Gehrig, J. Lecomte, A. V on Arnim, L. Benini, D. Scaramuzza, and A. Pantazi, “Neuromorphic optical flow and real-time implementation with event cameras,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2023, pp. 4129–4138

    2023

  50. [50]

    Spinnaker2: A large-scale neuromorphic system for event-based and asynchronous machine learning,

    H. A. Gonzalez, J. Huang, F. Kelber, K. K. Nazeer, T. Langer, C. Liu, M. Lohrmann, A. Rostami, M. Sch¨one, B. V oggingeret al., “Spinnaker2: A large-scale neuromorphic system for event-based and asynchronous machine learning,”arXiv preprint arXiv:2401.04491, 2024

    Pith/arXiv arXiv 2024

  51. [51]

    Using stochastic spiking neural networks on spinnaker to solve constraint satisfaction problems,

    G. A. Fonseca Guerra and S. B. Furber, “Using stochastic spiking neural networks on spinnaker to solve constraint satisfaction problems,” Frontiers in neuroscience, vol. 11, p. 714, 2017

    2017

  52. [52]

    Spiking analog vlsi neuron assemblies as constraint satisfaction problem solvers,

    J. Binas, G. Indiveri, and M. Pfeiffer, “Spiking analog vlsi neuron assemblies as constraint satisfaction problem solvers,” in2016 IEEE International Symposium on Circuits and Systems (ISCAS). IEEE, 2016, pp. 2094–2097

    2016

  53. [53]

    Solving constraint satisfaction problems with networks of spiking neurons,

    Z. Jonke, S. Habenschuss, and W. Maass, “Solving constraint satisfaction problems with networks of spiking neurons,”Frontiers in neuroscience, vol. 10, p. 118, 2016

    2016

  54. [54]

    Solving qubo on the loihi 2 neuromorphic processor,

    A. Pierro, P. Stratmann, G. A. F. Guerra, S. Risbud, T. Shea, A. R. Mangalore, and A. Wild, “Solving qubo on the loihi 2 neuromorphic processor,”arXiv preprint arXiv:2408.03076, 2024

    Pith/arXiv arXiv 2024

  55. [55]

    Solving quadratic unconstrained binary optimization with collaborative spiking neural networks,

    Y . Fang and A. S. Lele, “Solving quadratic unconstrained binary optimization with collaborative spiking neural networks,” in2022 IEEE International Conference on Rebooting Computing (ICRC). IEEE, 2022, pp. 84–88

    2022

  56. [56]

    Solving vertex cover via ising model on a neuromorphic processor,

    K. Corder, J. V . Monaco, and M. M. Vindiola, “Solving vertex cover via ising model on a neuromorphic processor,” in2018 IEEE International Symposium on Circuits and Systems (ISCAS). IEEE, 2018, pp. 1–5

    2018

  57. [57]

    Slack-free spiking neural network formulation for hypergraph minimum vertex cover,

    T. Nguyen, A.-D. Doan, T.-J. Chinet al., “Slack-free spiking neural network formulation for hypergraph minimum vertex cover,”Advances in Neural Information Processing Systems, vol. 37, pp. 66 410–66 430, 2024

    2024

  58. [58]

    Sparse coding by spiking neural networks: Convergence theory and computational results,

    P. T. P. Tang, T.-H. Lin, and M. Davies, “Sparse coding by spiking neural networks: Convergence theory and computational results,”arXiv preprint arXiv:1705.05475, 2017

    Pith/arXiv arXiv 2017

  59. [59]

    Posecnn: A convolu- tional neural network for 6d object pose estimation in cluttered scenes,

    Y . Xiang, T. Schmidt, V . Narayanan, and D. Fox, “Posecnn: A convolu- tional neural network for 6d object pose estimation in cluttered scenes,” arXiv preprint arXiv:1711.00199, 2017. JOURNAL OF LATEX CLASS FILES, VOL. 18, NO. 9, SEPTEMBER 2020 11

    Pith/arXiv arXiv 2017

  60. [60]

    Convposecnn: Dense con- volutional 6d object pose estimation,

    C. Capellen, M. Schwarz, and S. Behnke, “Convposecnn: Dense con- volutional 6d object pose estimation,”arXiv preprint arXiv:1912.07333, 2019

    Pith/arXiv arXiv 1912

  61. [61]

    Convolutional networks for real-time 6-dof camera relocalization,

    A. Kendall, M. Grimes, and R. Cipolla, “Convolutional networks for real-time 6-dof camera relocalization,”CoRR, 2015

    2015

  62. [62]

    Real-time seamless single shot 6d object pose prediction,

    B. Tekin, S. N. Sinha, and P. Fua, “Real-time seamless single shot 6d object pose prediction,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 292–301

    2018

  63. [63]

    Pvnet: Pixel- wise voting network for 6dof pose estimation,

    S. Peng, Y . Liu, Q. Huang, X. Zhou, and H. Bao, “Pvnet: Pixel- wise voting network for 6dof pose estimation,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019, pp. 4561–4570

    2019

  64. [64]

    Occlusion-robust object pose estimation with holistic representation,

    B. Chen, T.-J. Chin, and M. Klimavicius, “Occlusion-robust object pose estimation with holistic representation,” inProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2022, pp. 2929– 2939

    2022

  65. [65]

    Ep n p: An accurate o (n) solution to the p n p problem,

    V . Lepetit, F. Moreno-Noguer, and P. Fua, “Ep n p: An accurate o (n) solution to the p n p problem,”International journal of computer vision, vol. 81, no. 2, pp. 155–166, 2009

    2009

  66. [66]

    Edopt: Event-camera 6-dof dynamic object pose tracking,

    A. Glover, L. Gava, Z. Li, and C. Bartolozzi, “Edopt: Event-camera 6-dof dynamic object pose tracking,” in2024 IEEE International Con- ference on Robotics and Automation (ICRA). IEEE, 2024, pp. 18 200– 18 206

    2024

  67. [67]

    Line-based 6- dof object pose estimation and tracking with an event camera,

    Z. Liu, B. Guan, Y . Shang, Q. Yu, and L. Kneip, “Line-based 6- dof object pose estimation and tracking with an event camera,”IEEE Transactions on Image Processing, 2024

    2024

  68. [68]

    Test-time certifiable self-supervision to bridge the sim2real gap in event-based satellite pose estimation,

    M. Jawaid, R. Talak, Y . Latif, L. Carlone, and T.-J. Chin, “Test-time certifiable self-supervision to bridge the sim2real gap in event-based satellite pose estimation,” in2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2024, pp. 4534–4541

    2024

  69. [69]

    Evsat3d: Satellite pose estimation and 3d recon- struction with event camera,

    S. S. Malik, M. Moshrefizadeh, O. Tahri, X. Bai, E. Blasch, V . Sagan, and H. Aliakbarpour, “Evsat3d: Satellite pose estimation and 3d recon- struction with event camera,”IEEE Access, 2025

    2025

  70. [70]

    Model based training, detection and pose estimation of texture-less 3d objects in heavily cluttered scenes,

    S. Hinterstoisser, V . Lepetit, S. Ilic, S. Holzer, G. Bradski, K. Konolige, and N. Navab, “Model based training, detection and pose estimation of texture-less 3d objects in heavily cluttered scenes,” inAsian conference on computer vision. Springer, 2012, pp. 548–562

    2012

  71. [71]

    T-less: An rgb-d dataset for 6d pose estimation of texture-less objects,

    T. Hodan, P. Haluza, ˇS. Obdr ˇz´alek, J. Matas, M. Lourakis, and X. Zab- ulis, “T-less: An rgb-d dataset for 6d pose estimation of texture-less objects,” in2017 IEEE Winter Conference on Applications of Computer Vision (WACV). IEEE, 2017, pp. 880–888

    2017

  72. [72]

    Ycb-ev 1.1: Event-vision dataset for 6dof object pose estimation,

    P. Rojtberg and T. P ¨ollabauer, “Ycb-ev 1.1: Event-vision dataset for 6dof object pose estimation,” inEuropean Conference on Computer Vision. Springer, 2024, pp. 1–13

    2024

  73. [73]

    Rgb-de: Event camera calibration for fast 6-dof object tracking,

    E. Dubeau, M. Garon, B. Debaque, R. de Charette, and J.-F. Lalonde, “Rgb-de: Event camera calibration for fast 6-dof object tracking,” in 2020 IEEE International Symposium on Mixed and Augmented Reality (ISMAR). IEEE, 2020, pp. 127–135

    2020

  74. [74]

    E-pose: A large scale event camera dataset for object pose estimation,

    O. A. Hay, X. Huang, A. Ayyad, E. Sherif, R. Almadhoun, Y . Abdul- rahman, L. Seneviratne, A. Abusafieh, and Y . Zweiri, “E-pose: A large scale event camera dataset for object pose estimation,”Scientific Data, vol. 12, no. 1, p. 245, 2025

    2025

  75. [75]

    Event-rgb fusion for spacecraft pose estimation under harsh lighting,

    M. Jawaid, M. M ¨artens, and T.-J. Chin, “Event-rgb fusion for spacecraft pose estimation under harsh lighting,”Aerospace Science and Technol- ogy, p. 111039, 2025

    2025

  76. [76]

    Revisiting the p3p problem,

    Y . Ding, J. Yang, V . Larsson, C. Olsson, and K. ˚Astr¨om, “Revisiting the p3p problem,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 4872–4880

    2023

  77. [77]

    Methods for non-linear least squares problems,

    K. Madsen, H. B. Nielsen, and O. Tingleff, “Methods for non-linear least squares problems,” 2004

    2004

  78. [78]

    A method for the solution of certain non-linear problems in least squares,

    K. Levenberg, “A method for the solution of certain non-linear problems in least squares,”Quarterly of applied mathematics, vol. 2, no. 2, pp. 164–168, 1944

    1944

  79. [79]

    Epro-pnp: Generalized end-to-end probabilistic perspective-n-points for monocular object pose estimation,

    H. Chen, P. Wang, F. Wang, W. Tian, L. Xiong, and H. Li, “Epro-pnp: Generalized end-to-end probabilistic perspective-n-points for monocular object pose estimation,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 2781–2790

    2022

  80. [80]

    End-to-end learnable geometric vision by backpropagating pnp optimization,

    B. Chen, A. Parra, J. Cao, N. Li, and T.-J. Chin, “End-to-end learnable geometric vision by backpropagating pnp optimization,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion, 2020, pp. 8100–8109

    2020

Showing first 80 references.