Robot Manipulation on Calvin ABC -> D (Latency Measured)
0.45Average Path LengthRandom Skip
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Random Skip2026.02 | 0.45 | 22.6 | |
| HULC2026.02 | 0.67 | — | |
| FlexiDepth# Fine-tuned Parameters=19M, # Fine-tuned Steps=6.7 · 103, Training Cost (GPU Hour)=72026.02 | 1.65 | 27.6 | |
| SPIL2026.02 | 1.71 | — | |
| DeeR-VLA 3B# Fine-tuned Parameters=1.2B, # Fine-tuned Steps=9.2 · 104, Training Cost (GPU Hour)=1122026.02 | 2.82 | 19.3 | |
| DySL-VLA 3B# Fine-tuned Parameters=14M, # Fine-tuned Steps=6.7 · 103, Training Cost (GPU Hour)=72026.02 | 2.83 | 13.6 | |
| RoboFlamingo 3B2026.02 | 2.85 | 51 |