Hierarchical Control of Antiphase Synchronization in Biomorphic Spiking Neural Architectures for Locomotor Control

Artem Guba, Nikolay Kovalev, Alexey Zharinov, Vacheslav Demin, Viktor B. Kazantsev

TL;DR

This paper introduces a statistically validated framework for controlling "walking" rhythms (antiphase synchronization) in biomorphic spiking neural networks. By developing a new composite metric that outperforms standard phase-locking measures, the authors demonstrate that external stimulation type acts as a "master switch" for locomotor regimes, while internal synaptic weights serve to fine-tune stability.

What Problem Does This Paper Address?

Central Pattern Generators (CPGs) are the neural circuits responsible for rhythmic movements like walking, swimming, and breathing. While engineers have long used simplified mathematical oscillators to mimic these, building them with spiking neural networks (SNNs)—which more closely resemble real biology—is notoriously difficult.

The challenge lies in the "curse of dimensionality." Spiking models have dozens of parameters (synaptic weights, neuron properties, feedback loops) that interact nonlinearly. Finding the right settings to produce a stable, alternating "left-right" rhythm (antiphase synchronization) often feels like finding a needle in a haystack. This paper seeks to move beyond empirical "tuning" by systematically mapping how different inputs and network configurations drive these dynamic regimes in a closed-loop neuromuscular system.

Methods

The researchers built a biomorphic CPG based on the well-known Rybak architecture, which uses a "half-center oscillator" design where two neural modules reciprocally inhibit each other.

Key technical components include:

  • Neuron Model: They used the Izhikevich formalism, which balances biological realism with computational efficiency, allowing for large-scale simulations.
  • Closed-Loop System: The CPG isn't just "firing in a vacuum"; it's connected to a simulated muscle model and receives proprioceptive feedback (Ia, Ib, and II afferents) that signals muscle length, velocity, and force.
  • The Composite Metric ($S_{composite}$): Standard measures like the Phase Locking Value (PLV) only look at timing. The authors proposed a new heuristic metric that integrates:
  • Phase coherence (timing consistency).
  • Amplitude balance (ensuring both "legs" move with equal force).
  • Temporal stability (penalizing erratic bursting).
FIG. 1. Schematic diagram of the CPG architecture adapted from Rybak et al.31. The network consists of reciprocal inhibition circuits generating rhythmic bursting (CPG-N), afferent feedback pathways (Ia, Ib), and interneurons modulating afferent input (Ib-In, Ia-In, R-In).

Datasets

The study generated a massive, open-access dataset of over 10,000 simulation runs.

  • Varied Parameters: They systematically swept through different stimulation waveforms (Tonic/DC, Sine, Square) and synaptic weight configurations.
  • Validation Set: An expert-annotated subset of 7,000 segments was used to benchmark their new $S_{composite}$ metric against traditional methods.

Key Insights and Ideas

  • Control Hierarchy: The most significant finding is a clear hierarchy of control. The type of external drive (descending signals from the brain) determines the overall operating regime (e.g., whether the system oscillates at all), while synaptic weights within the spinal cord fine-tune the stability of that rhythm.
  • Tonic Drive is Superior: Counter-intuitively, a simple constant current (tonic drive) produced the most robust and stable walking rhythms. Periodic inputs (like sine waves) often conflicted with the network's internal rhythm, leading to nonlinear saturation and "glitchy" coordination.
  • The "Stabilizer" Synapse: Using Random Forest importance analysis, the authors identified reciprocal Ia interneuron connections as the most critical synaptic components for maintaining stable alternating patterns.
  • Regime Discrimination: The composite metric successfully separated four distinct behaviors: phase desynchronization (ataxia-like), in-phase synchronization (pathological co-contraction), no-oscillation (motor inhibition), and the desired anti-phase synchronization (locomotion).
The four primary dynamic regimes identified: (A) Desynchronized, (B) In-phase, (C) No-oscillation, and (D) Anti-phase synchronization.

Key Results

  • Metric Performance: The $S_{composite}$ metric achieved an AUC of 0.766 in detecting stable antiphase synchronization, a statistically significant improvement over the baseline PLV approach (AUC = 0.741).
  • Robustness Mapping: Statistical analysis (ANOVA) confirmed that stimulation type dominates regime determination. As shown in the violin plots below, tonic drive (DC) achieves much higher median synchronization values with a narrower spread compared to square or sine wave inputs.
Statistical validation shows that tonic (DC) drive provides the most stable and highest quality synchronization.

Limitations and Future Work

While the model is a significant step forward, the authors acknowledge several limitations:

  1. Fixed Descending Drive: The "brain" signals were static. Real biological systems use Spike-Timing-Dependent Plasticity (STDP) to adapt these signals over time for learning new gaits.
  2. Simplified Biomechanics: The muscle model is phenomenological and lacks details like tendon compliance, which can affect reflex latencies.
  3. Noise Modeling: They used additive Gaussian noise, which doesn't fully capture the stochastic nature of ion channels or hardware-level variability in neuromorphic chips.

Future directions include embedding these CPGs within reinforcement learning frameworks for adaptive gait modulation and deploying the architecture on specialized neuromorphic hardware like Intel's Loihi for energy-efficient, real-time robotic control.

Summary updated

Abstract

Central pattern generators (CPGs) provide a biologically inspired foundation for adaptive locomotor control, yet systematically characterizing their dynamic regimes in spiking neural networks remains challenging. This study introduces a framework for identifying and stabilizing antiphase synchronization in a biomorphic spiking CPG model embedded in a closed neuromuscular loop. A composite synchronization metric is developed, integrating phase coherence, amplitude balance, and temporal stability. Its performance is evaluated across systematically varied stimulation waveforms and synaptic configurations. Statistical analysis reveals that stimulation type dominates regime determination, with tonic drive yielding the most robust antiphase synchronization. Periodic inputs induce nonlinear saturation and context-dependent network routing. Although synaptic weights alone show limited discriminative power, they significantly modulate robustness within a given stimulation context, with reciprocal Ia interneuron connections emerging as key stabilizers. The proposed composite metric achieves robust antiphase detection, outperforming traditional phase-only measures while preserving biophysical interpretability. These findings establish a statistically validated, hierarchical modeling approach in which external drive defines the operating regime and targeted synaptic configurations fine-tune dynamic stability, providing actionable guidelines for closed-loop neuromechanical controllers and adaptive neuroprosthetic interfaces.
BibTeX
@misc{guba2026hierarchical,
  author = {Artem Guba and Nikolay Kovalev and Alexey Zharinov and Vacheslav Demin and Viktor B. Kazantsev},
  title = {{Hierarchical Control of Antiphase Synchronization in Biomorphic Spiking Neural Architectures for Locomotor Control}},
  year = {2026},
  eprint = {2611.22639},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2611.22639},
}
Figure