A Unified Neurocomputational Framework for Closed-Loop Motor Control and Sense of Agency

Abstract

Motor control relies on the closed-loop comparison of motor commands and sensory feedback to correct errors and adapt to perturbations. Relevant features must be selected from a rich stream of sensory inputs and bound to the appropriate motor commands. This process remains poorly understood. Closed-loop control is accompanied by the experience of causing the observed feedback, sense of agency (SoA). SoA grounds self-identification, and its impairment is associated with lower prosthesis acceptance and disorders like autism and schizophrenia. Like closed-loop motor control, it relies on comparing desired and observed action outcomes in fronto-parietal circuits. Yet, these two phenomena have been studied independently, leaving SoA without a functional meaning and disregarding subjective aspects in motor control models. We propose that SoA is the subjective correlate of selecting self-caused sensory features for closed-loop control. We tested this in a visuomotor task where SoA was manipulated through temporal delays and adaptation to spatial perturbations served as a proxy for closed-loop integration. Delays similarly modulated SoA and closed-loop integration, suggesting these may emerge from the same phenomenon. We modelled our results in a Bayesian framework in which self-causation probability is inferred from temporal congruence, jointly modulating SoA and the weight attributed to visual feedback.