Long-horizon video generation on RoboArena
17.66PSNRPersistWorld
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| PersistWorldDenoising steps=162026.07 | 17.66 | 0.0191 | 73.2 | 0.309 | 37.44 | 249.6 | |
| PersistWorldDenoising steps=82026.07 | 17.62 | 0.0193 | 73.1 | 0.314 | 40.64 | 306.3 | |
| PersistWorldDenoising steps=502026.07 | 17.58 | 0.0194 | 72.8 | 0.305 | 34.7 | 197.6 | |
| PersistWorldDenoising steps=322026.07 | 17.56 | 0.0196 | 72.9 | 0.307 | 35.32 | 208.2 | |
| PersistWorldDenoising steps=42026.07 | 17.01 | 0.0222 | 70.9 | 0.341 | 51.13 | 403 | |
| Step Forcing (Pre-train)Denoising steps=42026.07 | 16.92 | 0.022 | 71.3 | 0.295 | 28.28 | 197.3 | |
| Step Forcing (Pre-train 240k step)Denoising steps=42026.07 | 16.92 | 0.022 | 71.4 | 0.295 | 28.29 | 192 | |
| Step Forcing (Post-Train)Denoising steps=42026.07 | 16.79 | 0.023 | 71.5 | 0.293 | 28.54 | 203.83 | |
| Ctrl-WorldDenoising steps=162026.07 | 16.66 | 0.0232 | 71.2 | 0.324 | 38.16 | 229.8 | |
| Ctrl-WorldDenoising steps=322026.07 | 16.61 | 0.0237 | 70.6 | 0.322 | 35.38 | 190.8 | |
| Ctrl-WorldDenoising steps=82026.07 | 16.61 | 0.0232 | 71.2 | 0.333 | 43.83 | 317.2 | |
| Ctrl-WorldDenoising steps=502026.07 | 16.56 | 0.0242 | 70.4 | 0.321 | 34.8 | 190.2 | |
| Diffusion ForcingDenoising steps=42026.07 | 15.81 | 0.0292 | 68.3 | 0.327 | 33.18 | 242 | |
| Diffusion ForcingDenoising steps=82026.07 | 15.69 | 0.0302 | 67.8 | 0.327 | 29.98 | 210.54 | |
| Diffusion ForcingDenoising steps=162026.07 | 15.52 | 0.0313 | 67.2 | 0.331 | 29.16 | 190.21 | |
| Ctrl-WorldDenoising steps=42026.07 | 15.39 | 0.0303 | 66.4 | 0.392 | 69.17 | 529.7 | |
| Diffusion ForcingDenoising steps=322026.07 | 15.35 | 0.0323 | 66.1 | 0.338 | 30.07 | 178.96 | |
| Diffusion ForcingDenoising steps=502026.07 | 15.27 | 0.0329 | 65.3 | 0.343 | 31.81 | 186.54 | |
| IRASimDenoising steps=502026.07 | 13.34 | 0.0516 | 51.8 | 0.469 | 119.67 | 1,607 |