Video-level world modeling on robosuite Push
27.9346PSNR (10 steps)Orbisim w/o Decoupling
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Orbisim w/o DecouplingRollout Mode=Autoregressive, Decoupling=false2026.05 | 27.9346 | 19.951 | 0.1188 | 0.1799 | 689.1 | 0.8134 | |
| Orbisim (Final)Rollout Mode=Autoregressive2026.05 | 26.7105 | 19.9819 | 0.1078 | 0.1428 | 533.9 | 0.4468 | |
| Orbisim w/o Random SamplingRollout Mode=Autoregressive, Random Sampling=false2026.05 | 26.689 | 19.1119 | 0.1076 | 0.1669 | 531.2 | 0.5742 | |
| AdaWorldRollout Mode=Autoregressive2026.05 | 26.6647 | 12.8346 | 0.1183 | 0.3482 | 1,305.8 | 1.8597 | |
| Orbisim w/o Object-CentricRollout Mode=Autoregressive, Object-Centric=false2026.05 | 25.9373 | 19.7581 | 0.1123 | 0.1463 | 524.5 | 0.4687 | |
| Vid2WorldRollout Mode=Autoregressive2026.05 | 22.2014 | 17.8856 | 0.1312 | 0.2551 | 1,750.1 | 0.6754 |