Internal Dynamics Interact with Proprioceptive Feedback During Movement Execution in an RNN Model of Motor Cortex

Abstract

Proprioceptive feedback provides the information about the state of the body, which is critical in motor control. However, the contribution of proprioceptive feedback to motor cortical activity during voluntary movement execution is unclear. Here, we built an recurrent neural network model of motor cortex that receives proprioceptive feedback, and optimized it to control a virtual arm to perform a delayed-reach task. Simulated neural activity is similar to real data, indicating that our model captures the motor cortical dynamics. We further disrupted recurrent connectivity and proprioceptive feedback to dissect their contribution, and found that internal dynamics dominate in neural population activity, while proprioceptive feedback controls movement termination. Moreover, proprioceptive feedback improves the networks robustness against noisy initial conditions. We further investigated the relative importance of the components in proprioceptive feedback and found that the feedback of hand velocity contributes most to the similarity between simulation and real data. Finally, we show that our motor cortex model can be implemented in the sensorimotor system, demonstrating our models biological plausibility. In summary, motor command may arise from the intersection between recurrent dynamics in motor cortex and proprioceptive feedback.