Robot manipulation video generation on Self-curated Robot Manipulation Benchmark 1.0 (test)
24.24PSNROSCAR
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| OSCARHardware=NVIDIA GH200, Evaluation frame count=first 49 frames, Parameters=2B2026.06 | 24.24 | 0.846 | 0.094 | 0.015 | 7.08 | 15.07 | 0.096 | |
| Genie EnvisionerConditioning approach=explicit-geometry, Hardware=NVIDIA GH200, Evaluation frame count=first 49 frames2026.06 | 23.29 | 0.838 | 0.14 | 0.007 | 15.37 | 22.92 | 0.129 | |
| EnerVerse-ACConditioning approach=latent-action, Hardware=NVIDIA GH200, Evaluation frame count=first 49 frames2026.06 | 20.47 | 0.746 | 0.223 | 0.021 | 33.7 | 38.23 | 0.197 | |
| Ctrl-WorldConditioning approach=latent-action, Hardware=NVIDIA GH200, Evaluation frame count=first 49 frames2026.06 | 19.06 | 0.705 | 0.321 | 0.042 | 28.9 | 53.33 | 0.292 | |
| Kinema4DConditioning approach=explicit-geometry, Hardware=NVIDIA GH200, Evaluation frame count=first 49 frames, Parameters=14B2026.06 | 17.68 | 0.741 | 0.198 | 0.021 | 17.07 | 37.16 | 0.233 | |
| TesserActConditioning approach=text-only, Hardware=NVIDIA GH200, Evaluation frame count=first 49 frames2026.06 | 16.26 | 0.73 | 0.277 | 0.055 | 24.5 | 51.9 | 0.364 | |
| Cosmos-Predict2.5Conditioning approach=text-only, Hardware=NVIDIA GH200, Evaluation frame count=first 49 frames2026.06 | 14.78 | 0.563 | 0.37 | 0.022 | 18.01 | 47.59 | 0.435 | |
| IRASimConditioning approach=latent-action, Hardware=NVIDIA GH200, Evaluation frame count=first 49 frames2026.06 | 6.48 | 0.088 | 0.909 | 0.606 | 411.42 | 394.1 | 2.453 |