SeMOPO: Learning High-quality Model and Policy from Low-quality Offline Visual Datasets
The task-relevant observation reconstructions
(from top to bottom: raw observation, reconstruction, difference)
SeMOPO (Ours)
Cheetah
Cheetah
Walker
Walker
Hopper
Hopper
CarRacing
CarRacing
Humanoid
Humanoid
Separated Models + RS
Cheetah
Cheetah
Walker
Walker
Hopper
Hopper
Separated Models + RS + DRP
Cheetah
Cheetah
Walker
Walker
Hopper
Hopper