Code for "Synthesis of Biologically Realistic Human Motion Using Joint Torque Actuation", SIGGRAPH 2019 (https://arxiv.org/abs/1904.13041). We transform optimal control problems formulated with muscle actuators to equivalent problems using simple and efficient joint torque actuators, while retaining motion quality comparable to using muscles. We do so by learning two neural-net functions, state-dependent torque-limit function and metabolic-energy function from OpenSim muscle model by solving static redundancy problems. This project shows how to learn the two functions, and how to incorporate the learned neural networks into multi-CPU optimal control problems using Matlab, IPOPT and OpenSim.