作者: Josiah P. Hanna
关键词:
摘要: Transferring robotic control policies --- learned in simulation to physical robots is a promising alternative learning directly on the system. Unfortunately, often fail real world due inevitable discrepancies between and simulation. This thesis aims bridge gap reality by developing methods for grounding assessing how well policy will perform before it executed world. We discuss completed work towards simulation-transfer method of safe evaluation. then present directions future these areas.