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

Walker

Hopper

CarRacing

Humanoid

Separated Models + RS

Cheetah

Walker

Hopper

Separated Models + RS + DRP

Cheetah

Walker

Hopper