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

cs.ROApr 15, 2026

Neuromorphic Spiking Ring Attractor for Proprioceptive Joint-State Estimation

Authors: Federica Ferrari, Flavia Davidhi, Bernard Maacaron, Alberto Motta, Luuk van Keeken, Elisa Donati, Giacomo Indiveri, Chiara De Luca,

Organizations: Istituto Italiano di Tecnologia, Genoa, Italy · Department of Neurosurgery, University Hospital Zürich, Switzerland · Institute of Neuroinformatics, Univ. of Zurich and ETH Zurich, Switzerland · DIEEI, University of Catania Catania, Italy ·

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.

Figures & tables

Fig. 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 (

\theta_{0}

\theta_{l}

) and the middle out-of-bound angle

\theta_{m}^{*}

; gray area indicates the out-of-bound region.(d) The asymmetric profile in the bounded model obtained modulating the original profile through

a^{+}

and

a^{-}

. (e) DYNAP-SE symmetric profile, where number of connections increases with connection strength. (f) asymmetric profile in the DYNAP-SE implementation, where number of connections increase with the velocity intensity.

Fig. 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–constrained (red) models for wide (solid) and limited (dashed) motion ranges. Incorporating boundary conditions reduces drift and improves alignment with ground truth.

Fig. 3: Ring–attractor stability and velocity modulation on the DYNAP-SE neuromorphic processor. (a) Experimental setup: a 1- $\mathrm{s}$ 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 $\mathrm{s}$ ), pooled across ring populations; lower values indicate higher stability. (c) Mean bump angular velocity vs. number of velocity connections with weighted linear fit ( $w_{i}=1/\mathrm{SEM}_{i}^{2}$ ). For each connection count, angular velocity was computed by unwrapping the bump trace and fitting a line over 2.5–9.0 $\mathrm{s}$ . (d) Spike raster during 12 $\mathrm{s}$ baseline ( $v=0$ ). (e) Spike raster during velocity modulation showing baseline (0-1 $\mathrm{s}$ ), translation(2–9 $\mathrm{s}$ ) and recovery (9–12 $\mathrm{s}$ ). (f) Spike raster during “acceleration” test with stepwise addition and removal of velocity connections.

Continuous Attractor Neural Networks

See all 9 papers

A Spiking Ring-Attractor Architecture for Representation and Control of Revolute Robot Joints
Oct 19, 2026Aung Htet, Alejandro Rodriguez Jimenez, Ruidong Ma +1Continuous Attractor Neural Networks Robotic Control
Sheffield Hallam University, Sheffield, UK
Spiking Neuromorphic Control for Periodic Orbits Stabilization in Underactuated Robots: A Greedy Lyapunov-Based Approach
Oct 27, 2026Maksim Surov, Oleg SumenkovRobotic Control Robot Trajectory Tracking
The authors are with Mathematical Robotics Science Division, Sirius University of Science and Technology, Olympiyskiy Av. 1, Sirius, 354340 Russia
Hebbian Attractor Networks for Robot Locomotion
Mar 23, 2026Alexander Dittrich, Fuda van Diggelen, Dario FloreanoHebbian Learning Neural Network Training Dynamics
Laboratory of Intelligent Systems, EPFL, Switzerland

Oct 19, 2026J/K move · Enter open · S save

A Spiking Ring-Attractor Architecture for Representation and Control of Revolute Robot Joints

Aung Htet, Alejandro Rodriguez Jimenez, Ruidong Ma, Alessandro Di Nuovo

Sheffield Hallam University, Sheffield, UK

Continuous Attractor Neural Networks Robotic Control Spiking Neural Networks Recurrent Neural Networks

Open paper AskPDF arXiv

Robotic systems with revolute joints evolve on circular state spaces, yet most control architectures represent these states using Euclidean variables that ignore this underlying geometry. We propose a fully spiking neural control architecture in which ring attractors encode joint states, enforcing representations consistent with the topology of revolute systems. Homeostatic regulation computes goal-state mismatch directly over these distributed representations, and opponent gain-modulated populations convert this mismatch into controlled phase shifts of the activity bump, implementing set-point regulation as motion on the ring. We show that the bump dynamics follow an SO(2) geometry, and that gain asymmetries induce a signed phase velocity on the ring. To interface distributed activity with scalar regulation signals, we derive an analytic Fourier phase decoder and implement its signed components using push-pull spiking populations. For multi-joint systems, we demonstrate that local ring-based Fourier features compose into product representations that linearly decode SO(3) orientations through forward kinematics. Experiments in NEST validate the full SO(2) regulation mechanism and the spiking product-feature decoder for lift, pitch, and yaw. Robot experiments on a Tiago arm show that decoded ring dynamics generate joint velocity commands, producing directed motion with joint-dependent accuracy. Together, these results establish a neuromorphic control framework in which representation, regulation, and action are unified within a ring-attractor substrate aligned with the geometry of revolute systems.

Abstract

Continuous Attractor Neural Networks

Explore similar work

A Spiking Ring-Attractor Architecture for Representation and Control of Revolute Robot Joints

Spiking Neuromorphic Control for Periodic Orbits Stabilization in Underactuated Robots: A Greedy Lyapunov-Based Approach

Hebbian Attractor Networks for Robot Locomotion