New neural-inspired controller generalises modularity over lower limb tasks through internal models
Predictive neuromuscular models based on neural controllers are a powerful tool for testing assumptions on the underlying architecture of sensorimotor control and its associated neural activity. However, most current controllers suffer from lack of physiological plausibility and are generally task specific. We propose a new neural controller, called Internal Model-based Modular Controller (IMMC), where a hierarchical architecture organises generalizable modules in activation networks dedicated to different motion tasks. The architecture comprises a simple model of the mesencephalic locomotor region (MLR), which sends controlling signals that manage the activity of internal models (IMs). The IMs organise synergies, coordinated and stereotyped activity of multiple muscles, in task-specific networks. The resultant organisations allow the generalisation of this architecture to different lower limb motions. The IMMC was tested in Stand-To-Walk simulations (STW), where the MLR switches between two IMs that recombine five synergies to replicate the standing and walking tasks. The simulation kinematics, muscle activation patterns and ground reaction forces were generally consistent with experimental data. In addition, the controller can transition to slower and faster speeds by tuning a single controlling signal. The proposed architecture is a first step to develop neuromuscular models which integrate multiple motor behaviours in a unified controller.