Learning Optimization-based Control Policies Directly from Digital Twin Simulations | IEEE Conference Publication | IEEE Xplore