Using a Reinforcement Learning Controller to Overcome Simulator/Environment Discrepancies

doi 10.1109/icsmc.2001.973482