pith. sign in

arxiv: 2604.14021 · v1 · submitted 2026-04-15 · 💻 cs.RO

Neuromorphic Spiking Ring Attractor for Proprioceptive Joint-State Estimation

Federica Ferrari , Flavia Davidhi , Bernard Maacaron , Alberto Motta , Luuk van Keeken , Elisa Donati , Giacomo Indiveri , Chiara De Luca
show 1 more author
Chiara Bartolozzi
This is my paper

Pith reviewed 2026-05-10 13:07 UTC · model grok-4.3

classification 💻 cs.RO
keywords spiking neural networksring attractorneuromorphic hardwareproprioceptionjoint state estimationcontinuous attractorrobot controlactivity bump
0
0 comments X

The pith

A spiking ring attractor sustains a moving activity bump to represent robot joint angles on neuromorphic hardware.

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

The paper shows that a network of spiking neurons arranged in a ring can maintain an internal estimate of joint position without continuous external input. Local excitation and broad inhibition keep a localized group of neurons active as a stable bump, while velocity signals create temporary asymmetries in the synapses to shift the bump around the ring at a controllable speed. Boundary conditions at the ends of the joint range stop the bump from moving outside the mechanical limits, which reduces unwanted drift and improves tracking accuracy compared with models that allow free movement. The resulting circuit is compact and runs stably for many seconds while exhibiting an approximately linear relationship between the strength of the velocity modulation and the speed of the bump.

Core claim

A ring of spiking neurons connected with distance-dependent excitation and inhibition forms a persistent activity bump whose position encodes joint angle; asymmetric, velocity-modulated synaptic weights translate the bump along the ring, and hard boundary conditions at the joint limits confine the motion, yielding stable multi-second operation and a near-linear mapping from modulation amplitude to bump velocity.

What carries the argument

The self-sustaining activity bump in the spiking ring attractor, created by local excitation plus broad inhibition and translated by velocity-modulated synaptic asymmetries under boundary constraints.

If this is right

  • The network reproduces smooth trajectory tracking for robotic joints.
  • The bump remains stable and accurate near the mechanical joint limits.
  • Bump translation speed varies nearly linearly with the amplitude of the velocity-modulated synaptic input.
  • Drift is lower and accuracy is higher than in otherwise identical models without boundary conditions.
  • The entire circuit fits within the resource limits of current neuromorphic chips.

Where Pith is reading between the lines

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

  • Observing the bump's position over time could supply an implicit velocity estimate without an extra sensor.
  • The same ring structure might be replicated for multiple joints and wired together to form a full-body proprioceptive map.
  • Adding slow synaptic adaptation or homeostatic mechanisms could further extend stability beyond the demonstrated multi-second window.

Load-bearing premise

Local excitation combined with broad inhibition will keep the activity bump intact and able to move smoothly when synaptic inputs are made asymmetric by velocity signals, without collapse or uncontrolled drift under realistic spike noise and parameter variation.

What would settle it

A simulation or hardware run in which the activity bump collapses, fragments, or shows large random position drift over tens of seconds when realistic spike noise and small parameter mismatches are introduced would falsify the stability claim.

Figures

Figures reproduced from arXiv: 2604.14021 by Alberto Motta, Bernard Maacaron, Chiara Bartolozzi, Chiara De Luca, Elisa Donati, Federica Ferrari, Flavia Davidhi, Giacomo Indiveri, Luuk van Keeken.

Figure 1
Figure 1. Figure 1: Ring attractor architectures used in the simulation (with and without boundaries) and on DYNAP-SE. Top row: symmetric profiles across architectures, bottom row: asymmetric profiles. Connections are only shown for neuron on top. (a) The symmetric cosine-shaped profile in the original model. (b) The asymmetric sinusoidal profile in the original model. (c) Symmetric cosine-shaped profile with boundaries (θ0,θ… view at source ↗
Figure 2
Figure 2. Figure 2: Neural encoding and validation of angular trajectory tracking. (a) Tracking using instantaneous velocity input. Raster plot of spiking activity (black) and ground–truth joint angle (blue). Top: input velocity; left: initialization current; right: mean firing–rate profile within the highlighted region. (b) Tracking using target velocity commands only. Comparison between the unbounded (orange) and boundary–c… view at source ↗
Figure 3
Figure 3. Figure 3: Ring–attractor stability and velocity modulation on the DYNAP-SE neuromorphic processor. (a) Experimental setup: a 1-s Gaussian input is encoded by a LIF neuron and sent to the DYNAP-SE ring attractor; velocity connections are updated from the host computer. (b) Bump drift over time. Box plots show peak-position error across 0.5-s windows after stimulus offset (0–0.5 to 4.5–5.0 s), pooled across ring popul… view at source ↗
read the original abstract

Maintaining stable internal representations of continuous variables is fundamental for effective robotic control. Continuous attractor networks provide a biologically inspired mechanism for encoding such variables, yet neuromorphic realizations have rarely addressed proprioceptive estimation under resource constraints. This work introduces a spiking ring-attractor network representing a robot joint angle through self-sustaining population activity. Local excitation and broad inhibition support a stable activity bump, while velocity-modulated asymmetries drive its translation and boundary conditions confine motion within mechanical limits. The network reproduces smooth trajectory tracking and remains stable near joint limits, showing reduced drift and improved accuracy compared to unbounded models. Such compact hardware-compatible implementation preserves multi-second stability demonstrating a near-linear relationship between bump velocity and synaptic modulation.

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 introduces a spiking ring-attractor network for proprioceptive joint-state estimation. Local excitation and broad inhibition sustain a stable activity bump encoding joint angle; velocity-modulated synaptic asymmetries translate the bump; boundary conditions enforce mechanical limits. The work claims smooth trajectory tracking, multi-second stability near joint limits, reduced drift and improved accuracy relative to unbounded models, plus a near-linear relationship between bump velocity and synaptic modulation, all in a compact hardware-compatible implementation.

Significance. If the stability and linearity claims can be substantiated with quantitative evidence, the approach would supply a resource-efficient, biologically inspired mechanism for drift-resistant continuous state estimation on neuromorphic hardware, directly relevant to robotic proprioception under tight power and size constraints.

major comments (2)
  1. [Abstract] Abstract: the claims of 'reduced drift', 'improved accuracy', 'multi-second stability', and a 'near-linear relationship between bump velocity and synaptic modulation' are stated without any quantitative metrics, error bars, simulation parameters, comparison baselines, or hardware results, preventing assessment of effect size or validity.
  2. [Abstract] Abstract: the load-bearing assumption that local excitation plus broad inhibition, combined with velocity-modulated asymmetries and hard boundary conditions, yields a persistently stable, translatable bump is asserted but not tested against realistic spike noise or modest parameter mismatch; such conditions can break attractor symmetry and produce pinning, jitter, or edge leakage in spiking networks.
minor comments (1)
  1. The abstract interleaves methodological description with performance claims; separating these would improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We appreciate the referee's thorough review and constructive suggestions for improving our paper. We have made revisions to address the concerns raised regarding the abstract and the robustness of the proposed network. Our point-by-point responses are as follows.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claims of 'reduced drift', 'improved accuracy', 'multi-second stability', and a 'near-linear relationship between bump velocity and synaptic modulation' are stated without any quantitative metrics, error bars, simulation parameters, comparison baselines, or hardware results, preventing assessment of effect size or validity.

    Authors: We agree that the abstract would be strengthened by the inclusion of quantitative metrics. The manuscript body provides these details through simulation experiments, featuring error bars from multiple trials, comparisons against unbounded ring attractor models, and regression analysis for the velocity relationship. We have updated the abstract to include representative quantitative values drawn from these results, along with key simulation parameters and a note that the implementation is hardware-compatible but evaluated via simulation. revision: yes

  2. Referee: [Abstract] Abstract: the load-bearing assumption that local excitation plus broad inhibition, combined with velocity-modulated asymmetries and hard boundary conditions, yields a persistently stable, translatable bump is asserted but not tested against realistic spike noise or modest parameter mismatch; such conditions can break attractor symmetry and produce pinning, jitter, or edge leakage in spiking networks.

    Authors: This is a fair observation. The core simulations presented in the manuscript demonstrate the stability and translation of the activity bump under the specified network dynamics. To directly address potential concerns about noise and parameter sensitivity, we have added new simulation results in the revised manuscript. These include tests with realistic spike noise and small parameter mismatches, confirming that the bump maintains stability without pinning, jitter, or edge leakage under these conditions. revision: yes

Circularity Check

0 steps flagged

No circularity in derivation chain

full rationale

The provided abstract and context contain no equations, derivations, or explicit mathematical steps. The central claims (stable activity bump, near-linear velocity-modulation relationship, reduced drift) are presented as outcomes of a neuromorphic implementation and simulation rather than reductions to fitted inputs or self-citations. No load-bearing steps reduce by construction to prior results or definitions within the visible text. The reader's moderate circularity note is acknowledged but does not meet the threshold for flagging without quotable equations or self-referential fitting.

Axiom & Free-Parameter Ledger

3 free parameters · 2 axioms · 0 invented entities

The model rests on standard ring-attractor assumptions plus several tunable parameters whose values are not reported; no new physical entities are postulated.

free parameters (3)
  • local excitation strength
    Parameter controlling bump formation and stability, value not stated in abstract
  • broad inhibition width and strength
    Parameter setting the spatial extent of suppression, value not stated
  • velocity modulation factor
    Synaptic scaling coefficient that translates velocity into bump speed, value not stated
axioms (2)
  • domain assumption Spiking neurons with local excitation and broad inhibition can sustain a stable localized activity bump
    Invoked to justify the core ring-attractor mechanism
  • domain assumption Velocity signals can be translated into asymmetric synaptic modulation that shifts the bump without destabilizing it
    Core dynamic update rule assumed to hold in the spiking implementation

pith-pipeline@v0.9.0 · 5439 in / 1401 out tokens · 51971 ms · 2026-05-10T13:07:54.810528+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

25 extracted references · 25 canonical work pages

  1. [1]

    Head stabilization in a humanoid robot: models and implementations,

    E. Falotico, N. Cauli, P. Kryczka, K. Hashimoto, A. Berthoz, A. Takan- ishi, P. Dario, and C. Laschi, “Head stabilization in a humanoid robot: models and implementations,”Autonomous Robots, vol. 41, no. 2, pp. 349–365, 2017

    work page 2017

  2. [2]

    An anatomically constrained model for path integration in the bee brain,

    T. Stone, B. Webb, A. Adden, N. B. Weddig, A. Honkanen, R. Templin, W. Wcislo, L. Scimeca, E. Warrant, and S. Heinze, “An anatomically constrained model for path integration in the bee brain,”Current Biology, vol. 27, no. 20, pp. 3069–3085, 2017

    work page 2017

  3. [3]

    Neural dynamics for landmark orien- tation and angular path integration,

    J. D. Seelig and V . Jayaraman, “Neural dynamics for landmark orien- tation and angular path integration,”Nature, vol. 521, no. 7551, pp. 186–191, 2015

    work page 2015

  4. [4]

    Angular path integration by moving “hill of activity

    P. Song and X.-J. Wang, “Angular path integration by moving “hill of activity”: A spiking neuron model without recurrent excitation of the head-direction system,”Journal of Neuroscience, vol. 25, no. 4, p. 1002–1014, Jan. 2005

    work page 2005

  5. [5]

    Stability of working memory in continuous attractor networks under the control of short-term plas- ticity,

    A. Seeholzer, M. Deger, and W. Gerstner, “Stability of working memory in continuous attractor networks under the control of short-term plas- ticity,”PLOS Computational Biology, vol. 15, no. 4, p. e1006928, Apr. 2019

    work page 2019

  6. [6]

    Ring attractor dynamics in the drosophila central brain,

    S. S. Kim, H. Rouault, S. Druckmann, and V . Jayaraman, “Ring attractor dynamics in the drosophila central brain,”Science (New York, N.Y.), vol. 356, no. 6340, p. 849–853, May 2017

    work page 2017

  7. [7]

    Sensor fusion for robotic workspace state estimation,

    B. Olofsson, J. Antonsson, H. G. Kortier, B. Bernhardsson, A. Roberts- son, and R. Johansson, “Sensor fusion for robotic workspace state estimation,”IEEE/ASME Transactions on Mechatronics, vol. 21, no. 5, pp. 2236–2248, 2015

    work page 2015

  8. [8]

    Embodied neuromorphic intelligence,

    C. Bartolozzi, G. Indiveri, and E. Donati, “Embodied neuromorphic intelligence,”Nature communications, vol. 13, no. 1, p. 1024, 2022

    work page 2022

  9. [9]

    Neuromorphic electronic circuits for building autonomous cognitive systems,

    E. Chicca, F. Stefanini, C. Bartolozzi, and G. Indiveri, “Neuromorphic electronic circuits for building autonomous cognitive systems,”Proceed- ings of the IEEE, vol. 102, no. 9, pp. 1367–1388, 2014

    work page 2014

  10. [10]

    An on-chip spiking neural network for estima- tion of the head pose of the icub robot,

    R. Kreiser, A. Renner, V . R. Leite, B. Serhan, C. Bartolozzi, A. Glover, and Y . Sandamirskaya, “An on-chip spiking neural network for estima- tion of the head pose of the icub robot,”Frontiers in Neuroscience, vol. 14, p. 551, 2020

    work page 2020

  11. [11]

    ALTRO: A Fast Solver for Constrained Trajectory Optimization,

    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). Macau, China: IEEE Press, Nov. 2019, p. 4176–4181. [Online]. Available: https://doi.org/10.1109/IROS40897.2019.8967864

    work page doi:10.1109/iros40897.2019.8967864 2019

  12. [12]

    A scalable multicore architecture with heterogeneous memory structures for dynamic neu- romorphic asynchronous processors (dynaps),

    S. Moradi, N. Qiao, F. Stefanini, and G. Indiveri, “A scalable multicore architecture with heterogeneous memory structures for dynamic neu- romorphic asynchronous processors (dynaps),”IEEE Transactions on Biomedical Circuits and Systems, vol. 12, no. 1, p. 106–122, Feb. 2018

    work page 2018

  13. [13]

    Sensorimotor experience remaps visual input to a heading-direction network,

    Y . E. Fisher, J. Lu, I. D’Alessandro, and R. I. Wilson, “Sensorimotor experience remaps visual input to a heading-direction network,”Nature, vol. 576, no. 7785, pp. 121–125, 2019

    work page 2019

  14. [14]

    Closed-loop spiking control on a neuromorphic processor implemented on the icub,

    J. Zhao, N. Risi, M. Monforte, C. Bartolozzi, G. Indiveri, and E. Donati, “Closed-loop spiking control on a neuromorphic processor implemented on the icub,”IEEE Journal on Emerging and Selected Topics in Circuits and Systems, vol. 10, no. 4, p. 546–556, Dec. 2020

    work page 2020

  15. [15]

    Dynap-se2: a scalable multi-core dynamic neuromor- phic asynchronous spiking neural network processor,

    O. Richter, C. Wu, A. M. Whatley, G. K ¨ostinger, C. Nielsen, N. Qiao, and G. Indiveri, “Dynap-se2: a scalable multi-core dynamic neuromor- phic asynchronous spiking neural network processor,”Neuromorphic Computing and Engineering, vol. 4, no. 1, p. 014003, Jan. 2024

    work page 2024

  16. [16]

    Path integration and cognitive mapping in a continuous attractor neural network model,

    A. Samsonovich and B. L. McNaughton, “Path integration and cognitive mapping in a continuous attractor neural network model,”The Journal of Neuroscience, vol. 17, no. 15, p. 5900–5920, Aug. 1997

    work page 1997

  17. [17]

    Maintaining and updating accurate internal representations of continuous variables with a handful of neurons,

    M. Noorman, B. K. Hulse, V . Jayaraman, S. Romani, and A. M. Her- mundstad, “Maintaining and updating accurate internal representations of continuous variables with a handful of neurons,”Nature Neuroscience, vol. 27, no. 11, pp. 2207–2217, Nov. 2024

    work page 2024

  18. [18]

    Neuronal Population Coding of Movement Direction,

    A. P. Georgopoulos, A. B. Schwartz, and R. E. Kettner, “Neuronal Population Coding of Movement Direction,”Science, vol. 233, no. 4771, pp. 1416–1419, Sep. 1986

    work page 1986

  19. [19]

    Quantitative investigations of electrical nerve excitation treated as polarization. 1907,

    L. Lapicque, “Quantitative investigations of electrical nerve excitation treated as polarization. 1907,”Biological Cybernetics, vol. 97, no. 5-6, pp. 341–349, Dec. 2007

    work page 1907

  20. [20]

    Adaptive exponential integrate-and-fire model as an effective description of neuronal activity,

    R. Brette and W. Gerstner, “Adaptive exponential integrate-and-fire model as an effective description of neuronal activity,”Journal of Neurophysiology, vol. 94, no. 5, p. 3637–3642, Nov. 2005

    work page 2005

  21. [21]

    Neuromorphic electronic circuits for building autonomous cognitive systems,

    E. Chicca, F. Stefanini, C. Bartolozzi, and G. Indiveri, “Neuromorphic electronic circuits for building autonomous cognitive systems,”Proceed- ings of the IEEE, vol. 102, no. 9, p. 1367–1388, Sep. 2014

    work page 2014

  22. [22]

    S. R. Deiss, R. J. Douglas, and A. M. Whatley,A Pulse- Coded Communications Infrastructure for Neuromorphic Systems. The MIT Press, Nov. 1998, p. 157–178. [Online]. Avail- able: https://direct.mit.edu/books/book/2001/chapter/54534/A-Pulse- Coded-Communications-Infrastructure-for

    work page 1998

  23. [23]

    Brian 2, an intuitive and efficient neural simulator,

    M. Stimberg, R. Brette, and D. F. Goodman, “Brian 2, an intuitive and efficient neural simulator,”eLife, vol. 8, p. e47314, Aug. 2019

    work page 2019

  24. [24]

    Mujoco: A physics engine for model-based control,

    E. Todorov, T. Erez, and Y . Tassa, “Mujoco: A physics engine for model-based control,” in2012 IEEE/RSJ International Conference on Intelligent Robots and Systems. IEEE, 2012, pp. 5026–5033

    work page 2012

  25. [25]

    icub: The not-yet-finished story of building a robot child,

    L. Natale, C. Bartolozzi, D. Pucci, A. Wykowska, and G. Metta, “icub: The not-yet-finished story of building a robot child,”Science Robotics, vol. 2, no. 13, p. eaaq1026, 2017

    work page 2017