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

Abstract

AbstractPredictive neuromuscular models are a powerful tool for testing assumptions on the underlying architecture of sensorimotor control and its associated neural activity. These models can test hypotheses that conventional methods of assessment cannot evaluate directly. However, current models are generally task-specific and mapping completely the skill space of a motion requires the tuning of all the parameters of the system. We here propose a modular model for Posture and Locomotion (MPL model), where a hierarchical architecture organizes modules in specific activation networks to accomplish motion tasks. A higher control layer, represented by a hypothetical mesencephalic locomotor region (MLR), sends controlling signals that manage motions and maps the skill space. The switch of motions is reflected by the activation of internal models (IMs). The IMs organize modules called synergies, that are coactivated sensory responses mapping multiple muscles, to display different motor behaviours. This architecture was tested in stand-to-walk (STW) simulations, where two IMs recombine five synergies to replicate ‘stand’ and ‘walk’. The model was successful in replicating a STW transition in a single simulation. The kinematics, muscle activation patterns and ground reaction forces during walking are consistent with experimental data. The model is also able to transition to slower and faster speeds by tuning the controlling signal once steady gait is reached. The proposed architecture is expected to be a first step to create neuromuscular models which integrate multiple motor behaviours in a unified controller.