“When environment parameters were changed mid-episode in simulation, the control policy initially performed worse but then adapted and returned to its normal performance level.”