A novel modular architecture for a neural controller architecture for predictive simulations of stand-to-walk motions

Abstract

Computational models of neural control are a powerful tool for evaluating principles of motor control that cannot be tested directly in conventional experimental settings. Current gait controllers struggle with physiological grounding and can replicate a small set of discrete behaviours. Objectivehere we propose a new neural controller, the Internal Model-based Modular Controller, that can transition between multiple lower-limb motor tasks within a single simulation. The architecture comprises a simplified model of the Mesencephalic Locomotor Region, which activates different internal models. These internal models organise functional muscle synergies into task-specific networks that generate coordinated activity across multiple muscles. Resultsthe controller performs Stand-To-Walk-To-Stand and Stand-To-Backward Walking-To-Stand transitions and modulates gait speed within a simulation by adjusting a few control signals. The observed simulated biomechanics and muscle activation patterns are generally consistent with experimental observations. ConclusionsOur results show that the proposed modular architecture represents a plausible mechanism for producing heterogeneous motor behaviours. IMPACT STATEMENTThis study presents a modular neural controller, which employs internal models and functional muscle synergies to reproduce heterogeneous legged behaviours within a single simulation framework.