A physiologically inspired hybrid CPG/Reflex controller for cycling simulations that generalizes to walking

Predictive simulations based on explicit, physiologically inspired, control policies, can be used to test theories on motor control and to evaluate the effect of interventions on the different components of control. Several control architectures have been proposed for simulating locomotor tasks, based on fully feedback, reflex-based, controllers, or on feedforward architectures mimicking the Central Pattern Generators. Hybrid architectures integrating both feedback and feedforward components represent a viable alternative to fully feedback or feedforward controllers. Current literature on controller-based simulations almost exclusively presents task-specific controllers that do not generalize across different tasks. The task-specificity of current controllers limits the generalizability of the neurophysiological principles behind such controllers. Here we propose a hybrid controller for predictive simulations of cycling where the feedforward component is based on a well-known theoretical model, the Unit Burst Generation model, and the feedback component includes a limited set of reflex pathways, expected to be active during steady cycling. We show that this controller can simulate physiological stationary cycling patterns at different desired speeds and seat heights. We also show that the controller can generalize to walking behaviors by just adding an additional control component for accounting balance needs. The controller here proposed, although simple in design, represent an instance of physiologically inspired generalizable controller for cyclical lower limb tasks.Author summary: Predictive simulations allow to synthesize human movements and their associated biomechanical quantities without using experimental data. Such simulations can help understand how humans may perform movement tasks in different scenarios and could provide useful information in fields like rehabilitation. One of the methodologies used for developing predictive simulations is based on creating models of the neural control architectures that generate the activation of the muscles. In this work we propose the first neural control architecture for cycling behaviors. We show that our architecture can be used to develop predictive simulations of cycling at different speeds and that the associated biomechanical quantitites are consistent with experimental data. Moreover, we show that our control architecture can also replicate walking behaviors with minimal modifications.