Long-horizon Manipulation on RLBench (Block Stacking Subset)
88Bridge Between Towers Success RateRoboStream-32B
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| RoboStream-32BModel Scale=32B, Zero-shot=true, Number of episodes=252026.03 | 88 | 84 | 68 | 100 | 76 | 72 | 96 | 76 | 82.5 | |
| RoboStream-235BModel Scale=235B, Zero-shot=true, Number of episodes=252026.03 | 88 | 92 | 96 | 100 | 88 | 80 | 96 | 84 | 90.5 | |
| RoboStream-8BModel Scale=8B, Zero-shot=true, Number of episodes=252026.03 | 76 | 40 | 16 | 100 | 24 | 60 | 96 | 52 | 58 | |
| SoFarNumber of episodes=252026.03 | 52 | 4 | 0 | 100 | 4 | 4 | 60 | 0 | 28 | |
| VoxPoserNumber of episodes=252026.03 | 8 | 4 | 0 | 96 | 4 | 4 | 96 | 0 | 26.5 |