Tunable Thresholds and Frequency Encoding in a Spiking NOD Controller

Ian Xul Belaustegui, Alessio Franci, Naomi Ehrich Leonard · 2025

Read Tunable Thresholds and Frequency Encoding in a Spiking NOD Controller by Ian Xul Belaustegui, Alessio Franci, Naomi Ehrich Leonard (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-01878.txt.

Introduction

Ian Xul Belaustegui, Alessio Franci, and Naomi Ehrich Leonard extend a spiking opinion model into a tunable controller in a 2025 paper. The model borrows from neurons: views are held as spikes in continuous time, letting a system switch choices fast when conditions change. The design goal is agile group decisions.

The authors find when steady spiking begins, giving conditions for the threshold that triggers it. They show how base sensitivity tunes that threshold and how spike rate above it encodes input strength. The balance between staying firm and reacting fast can thus be set by design. They close by casting the model as a neuromorphic control block for wider networks.

Catalog record

CREATORIan Xul Belaustegui, Alessio Franci, Naomi Ehrich Leonard
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE35 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-01878

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Tunable Thresholds and Frequency Encoding in a Spiking NOD Controller Ian Xul Belaustegui, Alessio Franci, Naomi Ehrich Leonard Abstract Spiking Nonlinear Opinion Dynamics (S-NOD) is an excitable decision-making model inspired by the spiking dynamics of neurons. S-NOD enables the design of agile decision-making that can rapidly switch between decision options in response to a changing environment. In S-NOD, decisions are represented by discrete opinion spikes that evolve in continuous time. Here, we extend previous analysis of S-NOD and explore its potential as a nonlinear controller with a tunable balance between robustness and responsiveness to input. We identify and provide necessary conditions for the bifurcation that determines the onset of periodic opinion spiking. We leverage this analysis to characterize the tunability of the input-output threshold for opinion spiking as a function of the model basal sensitivity and the tunable dependence of opinion spiking frequency on input magnitude above the threshold. We conclude with a discussion of S-NOD as a new neuromorphic control block and its extension to distributed spiking controllers. I Introduction Decision-making—the integration of sensory information and internal objectives to select an action—is a fundamental process in biological systems, operating across all levels of organization, from molecular networks [ 1 ] to neural populations [ 2 ] and collective behaviors [ 3 ] . Modeling biological decision-making is crucial not only for understanding its underlying mechanisms but also for engineering bio-inspired autonomous systems. Traditional models of opinion dynamics often struggle to capture the robust yet flexible decision-making behavior observed in biological systems. The Nonlinear Opinion Dynamics (NOD) model [ 4 , 5 ] shows how balancing negative and positive feedback can lead to robust, fast, and flexible decision-making through an organizing pitchfork bifurcation. However, NOD may not always be able to switch between decision options with sufficient agility in rapidly changing environments. The Spiking-NOD (S-NOD) model [ 6 ] addresses this limitation by making NOD excitable, in the sense that it has a roughly linear response to input below a threshold, but can produce large excursions in phase-space ( spikes ) for input above the threshold [ 7 ] . Excitability is achieved through adaptive slow negative feedback that brings the NOD state back to its neutral, ultra-sensitive state each time a decision is made, resulting in excitable opinion spikes . As an excitable system, S-NOD exhibits a form of controlled instability, where positive feedback leads to large and fast deviations away from equilibrium while slow negative feedback ensures regulation toward it. This mechanism has been successfully applied to robot navigation in crowded environments, ensuring responsive and decisive behavior [ 6 ] . The use of excitability for control is central to neuromorphic engineering [ 8 ] , with applications demonstrated in tasks such as the rhythmic control of a pendulum [ 9 ] . Fig. 1 , adapted from [ 6 ] , illustrates the agility of the S-NOD controller. Here, we extend the analysis in [ 6 ] to rigorously characterize two key tunable features of S-NOD useful for control: the spiking threshold and the response of spiking frequency to input magnitude. Section II introduces S-NOD and the notation for our analysis. In Section III , we study the dependence of S-NOD dynamics on exogenous input and prove that spiking emerges through a Hopf bifurcation at a tunable input threshold. In Section IV , we study deadlock breaking where two symmetric spiking limit cycles appear in the absence of input as the system undergoes a symmetry-breaking pitchfork bifurcation at the critical basal sensitivity. In Section V , we use geometric arguments to characterize how the spiking threshold and frequency response to input can be tuned as a function of model parameters. This paper thus provides an in-depth dynamical systems characterization of S-NOD as a tunable and agile spiking control element. Fig. 1: Figure from [ 6 ] shows S-NOD (pink) applied to control a planar robot in a social navigation scenario. NOD (blue) is shown for comparison. The robot aims to reach a goal marked by a star while avoiding collision with an unpredictable oncoming mover (black). The NOD controller is slow to respond to an abrupt change in the direction of the oncoming mover, but the S-NOD controller responds with agility and avoids colliding. II Spiking Nonlinear Opinion Dynamics S-NOD is governed by the differential equations \displaystyle\dot{z} \displaystyle=-d\,z+\tanh((kz^{2}+\mu_{0}-s)az+b) (1) \displaystyle\dot{s} \displaystyle=\varepsilon(-s+k_{s}z^{4})\,. The state z(t)\in\mathbb{R} is the agent’s opinion at time t . For decision-making between two mutually exclusive options, we interpret z(t)>0 ( z(t)<0 ) as an opinion in favor of option 1 (option 2). |z(t)| is the strength of the opinion and z(t)=0 is a neutral opinion (indecision). s(t)\in\mathbb{R} is the agent’s slow recovery state at time t . d\in\mathbb{R}_{+} is the gain of negative feedback opinion-damping term -d\,z . \mu_{0}\in\mathbb{R}_{+} sets the basal level of the agent’s sensitivity 1 1 1 In [ 6 ] \mu , \mu_{0} are u , u_{0} and called “attention” and “basal attention.” \mu to positive feedback term az , where \mu=kz^{2}+\mu_{0}-s . The agent’s sensitivity \mu increases with gain k\in\mathbb{R}_{+} as the magnitude of the agent’s opinion increases. At a much slower time-scale, \mu decreases as the slow recovery variable s increases. This creates a balance of fast positive and slow negative feedback. \varepsilon\in\mathbb{R}_{+} , \varepsilon\ll 1 sets the timescale separation between the dynamics of z and s : z is fast and s is slow. \mu_{o},a,k\in\mathbb{R}_{+} tune the positive feedback gain. k_{s}\in\mathbb{R}_{+} tunes the slow negative feedback gain. b\in\mathbb{R} is an exogenous input. In applications b would integrate various task-relevant signals, and represent evidence for a decision in the positive or negative z direction. In the following analysis we consider b to be a constant parameter. For our results to apply directly in situations where b is a time-varying signal, we would require it to have a timescale slower than the dynamics ( 1 ). We first establish some basic results needed to analyze the onset of opinion spiking as either input b (in Section III ) or tuning parameter \mu_{0} increases (in Section IV ). Remark II.1 If the point (\bar{z},\bar{s})\in\mathbb{R}^{2} is an equilibrium of ( 1 ) then \bar{s}=k_{s}z^{4} , and \bar{z} solves the equation h_{b}(\bar{z}):=-d\bar{z}+\tanh(a\bar{z}(-k_{s}\bar{z}^{4}+k\bar{z}^{2}+\mu_{0})+b)=0. (2) We will say \bar{z}\in\mathbb{R} is an equilibrium if it satisfies ( 2 ). The stability of equilibria of two-dimensional systems is fully determined by the trace and determinant of the Jacobian evaluated at equilibrium. Let \varphi(z,s):=az(kz^{2}+\mu_{0}-s)+b and \psi_{b}(z):=\varphi(z,k_{s}z^{4})=az(-k_{s}z^{4}+kz^{2}+\mu_{0})+b . Letting \tanh^{\prime}(x)=\frac{d\tanh(x)}{dx}=1-\tanh^{2}(x) , the Jacobian of ( 1 ) is \displaystyle J(z,s)\!:=\!\begin{pmatrix}-d+\tanh^{\prime}(\varphi)\frac{\partial\varphi}{\partial z}&\tanh^{\prime}(\varphi)\frac{\partial\varphi}{\partial s}\\ 4\varepsilon k_{s}z^{3}&-\varepsilon\end{pmatrix}\,. (3) Remark II.2 If \bar{z}\in\mathbb{R} is an equilibrium then \tanh^{\prime}(\varphi(\bar{z},k_{s}\bar{z}^{4}))=1-d^{2}\bar{z}^{2} . For notation purposes let J_{z}:=J(z,k_{s}z^{4}) be the Jacobian evaluated over the s -nullcline. The following result follows from elementary algebraic manipulations. Lemma II.3 If \bar{z} is an equilibrium, then \displaystyle\text{tr}(J_{\bar{z}}) \displaystyle=\underbrace{ad^{2}k_{s}}_{c_{3}}\bar{z}^{6}-\underbrace{(3ad^{2}k+ak_{s})}_{c_{2}}\bar{z}^{4} \displaystyle+\underbrace{(3ak-ad^{2}\mu_{0})}_{c_{1}}\bar{z}^{2}+\underbrace{a\mu_{0}-d}_{c_{0}}-\varepsilon (4) \displaystyle\frac{\det(J_{\bar{z}})}{\varepsilon} \displaystyle=-\underbrace{5ad^{2}k_{s}}_{\hat{c}_{3}=5c_{3}}\bar{z}^{6}+\underbrace{(3ad^{2}k+5ak_{s})}_{\hat{c}_{2}=c_{2}+4ak_{s}}\bar{z}^{4} \displaystyle-\underbrace{(3ak-ad^{2}\mu_{0})}_{\hat{c}_{1}=c_{1}}\bar{z}^{2}-\underbrace{(a\mu_{0}-d)}_{\hat{c}_{0}=c_{0}}. (5) Finding the roots of ( 4 ) and ( 5 ) reduces to solving two third-degree polynomials, namely, q_{tr}(z):=\text{tr}(J_{\sqrt{z}}) and q_{det}(z):=\det(J_{\sqrt{z}})/\varepsilon . III Spiking from increasing b Fig. 2: Bifurcation diagram for ( 1 ) with respect to input ( b ) projected onto the opinion ( z ) axis to reduce from three dimensions to two. Solid lines indicate stable equilibria, dashed lines indicate unstable

Read the complete text