Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Neuromorphic Architecture for Scalable Event-Based Control

Yongkang Huo, Fulvio Forni, Rodolphe Sepulchre

TL;DR

The paper tackles the challenge of unifying discrete decision logic with continuous regulation in scalable control systems. It introduces rebound excitability and RWTA as a single, event-based substrate that can generate rhythms and perform decisions across multiple scales, demonstrated on a five-link robotic snake. The approach provides modular, robust control without the need for glue logic between heterogeneous design domains, leveraging hierarchical RWTA motifs from cellular to system level. The work showcases the potential of a fully neuromorphic, embodied computation framework for scalable, adaptive locomotion and beyond, while outlining future directions in learning and hardware realization.

Abstract

This paper introduces the ``rebound Winner-Take-All (RWTA)" motif as the basic element of a scalable neuromorphic control architecture. From the cellular level to the system level, the resulting architecture combines the reliability of discrete computation and the tunability of continuous regulation: it inherits the discrete computation capabilities of winner-take-all state machines and the continuous tuning capabilities of excitable biophysical circuits. The proposed event-based framework addresses continuous rhythmic generation and discrete decision-making in a unified physical modeling language. We illustrate the versatility, robustness, and modularity of the architecture through the nervous system design of a snake robot.

A Neuromorphic Architecture for Scalable Event-Based Control

TL;DR

The paper tackles the challenge of unifying discrete decision logic with continuous regulation in scalable control systems. It introduces rebound excitability and RWTA as a single, event-based substrate that can generate rhythms and perform decisions across multiple scales, demonstrated on a five-link robotic snake. The approach provides modular, robust control without the need for glue logic between heterogeneous design domains, leveraging hierarchical RWTA motifs from cellular to system level. The work showcases the potential of a fully neuromorphic, embodied computation framework for scalable, adaptive locomotion and beyond, while outlining future directions in learning and hardware realization.

Abstract

This paper introduces the ``rebound Winner-Take-All (RWTA)" motif as the basic element of a scalable neuromorphic control architecture. From the cellular level to the system level, the resulting architecture combines the reliability of discrete computation and the tunability of continuous regulation: it inherits the discrete computation capabilities of winner-take-all state machines and the continuous tuning capabilities of excitable biophysical circuits. The proposed event-based framework addresses continuous rhythmic generation and discrete decision-making in a unified physical modeling language. We illustrate the versatility, robustness, and modularity of the architecture through the nervous system design of a snake robot.

Paper Structure

This paper contains 18 sections, 3 equations, 20 figures, 1 table.

Table of Contents

  1. Introduction
  2. Cellular Level: Rebound Excitability, Neuronal Models, and Synapses.
  3. Rebound Excitability
  4. Rebound Spiking and Bursting
  5. Inhibitory and Excitatory Synapses
  6. Functional Level: Decision Making
  7. Functional Level: Pattern Generation
  8. Functional Level: Interaction of the RWTA Network
  9. Multiscale States and Network Hierarchy
  10. Neuromorphic control of a five-link robotic snake
  11. Snake mechanics, muscle actuation, and neuromorphic controller basics
  12. Muscle Actuation Layer
  13. Coordination Layer
  14. Supervisory Control Layer
  15. Discussion
  16. ...and 3 more sections

Figures (20)

  • Figure 1: Design requirements for each design level in each task(layer).
  • Figure 2: Circuit model of neuron
  • Figure 3: Rebound neurons.
  • Figure 4: Two voltage trajectories, A and B, of the same neuron for different values of $\alpha_{us}^{+}$. The input current is an inhibitory step from -1.5 to -5 with duration 50 ms. Trajectory A (B) corresponds to a smaller (larger) $\alpha_{us}^{+}$.
  • Figure 5: Decision Making Motifs. Fast synapses ($\tau=0.5$) are represented with blue, slow synapses ($\tau=50$) are represented with red. Arrow for excitatory connection and circle for inhibitory connections. Dashed arrows denote input and output connections.
  • ...and 15 more figures