muSim: A goal-driven framework for elucidating the neural control of movement through musculoskeletal modeling

How does the motor cortex (MC) produce purposeful and generalizable movements with the complex musculoskeletal system in a dynamic environment? To elucidate the underlying neural dynamics, we use a goal-driven approach to model MC by considering its goal as a controller driving the musculoskeletal system through desired states to achieve movement. Specifically, we formulate a model of MC as a recurrent neural network (RNN) controller producing muscle commands while receiving sensory feedback from biologically accurate musculoskeletal models. Given this real-time simulated feedback implemented in advanced physics simulation engines, we use deep reinforcement learning to train the RNN to execute desired movements under specified neural and musculoskeletal constraints. For general use, we provide a modular computational framework that allows the flexible integration of user-defined musculoskeletal models, training algorithms, tasks and constraints. We also provide a combination of modules to analyze and quantify the dynamical alignment and similarity of the trained RNN with the recorded neural data on the population and single-unit level. Using these modules, we find that the activity of the trained RNN can accurately decode experimentally recorded neural population dynamics and single-unit MC activity, while generalizing well to testing conditions significantly different from training. Finally, we also provide perturbation modules to generate insights about neural dynamics for perturbed conditions different from training, and show that this framework unveils computational principles of how such neural dynamics enable flexible control of movement.