CO-CONTRACTION AS AN ADAPTIVE RESPONSE TO TASK DYNAMICS AND UNCERTAINTY

The central nervous system (CNS) can rely on different sensorimotor control strategies to regulate movements in uncertain and dynamic environments. Using sensory information, it can implement feedback control to perform task-relevant corrections but is limited by sensorimotor delays and noise (Franklin & Wolpert, 2011). Alternatively, the CNS can use statistical regularities from past observations to anticipate disturbances through feedforward control, reflected in preparatory changes of limb mechanical impedance via muscle co-contraction (Leib et al., 2024). Recent computational approaches suggest that skilled sensorimotor control stems from a context-optimal combination of such feedback and feedforward control (Yeo et al., 2016; Berret & Jean, 2020; Berret et al., 2021).

This study investigates how humans modulate the contributions of feedforward and feedback control depending on task dynamics and uncertainty. Sixteen participants performed a pointing task using a wrist exoskeleton equipped with a force sensor, while surface electromyographic activity was recorded from flexors and extensors muscles. Participants first practiced a baseline condition without perturbation, followed by three conditions with mechanical perturbation at 30%, 60%, or 90% of movement amplitude (50% probability per block).

Consistent with theoretical predictions (Berret et al., 2024), muscle co-contraction and grip force increased in uncertain conditions compared to baseline, including in unperturbed trials. Furthermore, co-contraction increased as perturbations occurred closer to the target, indicating that humans adapt their co-contraction level to task demands. When the mechanical disturbance occurs late in the movement, feedback control is less effective due to inherent delays and noise, and the CNS increasingly relies on feedforward control.