Robotic Manipulation on RoboTwin 2.0 (Average)
92.62Success RateMECo-WAM
Evaluation results
| Method | Links | ||
|---|---|---|---|
| MECo-WAMEmbodied-policy pre-training=false2026.07 | 92.62 | — | |
| LingBot-VALatency (ms)=843.57 ± 11.55, VRAM (GiB)=12.97, Action-denoising steps=10, Video-denoising steps=52026.06 | 92.3 | — | |
| LingBot-VAEmbodied-policy pre-training=true2026.07 | 92.2 | — | |
| Fast-WAMEmbodied-policy pre-training=false2026.07 | 91.83 | — | |
| Fast-WAMLatency (ms)=425.53 ± 6.01, VRAM (GiB)=11.55, Action-denoising steps=102026.06 | 91.8 | — | |
| WAM4DLatency (ms)=525.43 ± 5.64, VRAM (GiB)=9.71, Action-denoising steps=10, Video-denoising steps=52026.06 | 91.8 | — | |
| LingBot-VA*Nv=25, Na=502026.06 | 91.25 | 1 | |
| MotusLatency (ms)=1516.30 ± 10.64, VRAM (GiB)=11.55, Action-denoising steps=102026.06 | 87.9 | — | |
| MotusEmbodied-policy pre-training=true2026.07 | 87.84 | — | |
| Motus2026.06 | 87.8 | — | |
| Flash-WAMNv=1, Na=22026.06 | 85.54 | 19 | |
| Flash-WAMNv=1, Na=12026.06 | 81.41 | 23.3 | |
| π0.52026.06 | 79.8 | — | |
| π0.5Latency (ms)=72.03 ± 0.06, VRAM (GiB)=8.45, Action-denoising steps=102026.06 | 79.8 | — | |
| π0.5Embodied-policy pre-training=true2026.07 | 79.75 | — | |
| LingBot-VA + Video-only LCMNv=1, Na=22026.06 | 78.79 | 19 | |
| LingBot-VA + DMD2Nv=1, Na=22026.06 | 78.74 | 19 | |
| LingBot-VA + Video-only LCMNv=1, Na=12026.06 | 73.68 | 23.3 | |
| X-VLAEmbodied-policy pre-training=true2026.07 | 72.82 | — | |
| X-VLA2026.06 | 72.8 | — | |
| π02026.06 | 62.2 | — | |
| π0Latency (ms)=64.16 ± 0.06, VRAM (GiB)=8.45, Action-denoising steps=102026.06 | 62.2 | — | |
| LingBot-VA + DMD2Nv=1, Na=12026.06 | 50.56 | 23.3 | |
| LingBot-VA + Naive Joint LCMNv=1, Na=12026.06 | 36.32 | 23.3 | |
| LingBot-VA + Naive Joint LCMNv=1, Na=22026.06 | 23.97 | 19 |