Long-horizon Robotic Instruction Following on Libero long
100Soup and Sauce in Basket Success Rateπ0.5 + TACO
Evaluation results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| π0.5 + TACOaction sampling temperature=12026.05 | 100 | 96 | 98 | 100 | 98 | 100 | 92 | 100 | 86 | 96 | 96.6 | |
| CAPSaction sampling temperature=1, base policy=π0.52026.05 | 100 | 98 | 99 | 99 | 98 | 100 | 97 | 99 | 91 | 95 | 97.6 | |
| π0.5action sampling temperature=12026.05 | 98 | 100 | 98 | 98 | 98 | 100 | 96 | 94 | 68 | 98 | 94.8 | |
| Robomonkey*action sampling temperature=12026.05 | 59 | 79 | 58 | 37 | 55 | 86 | 59 | 62 | 26 | 44 | 56.5 | |
| OpenVLA*action sampling temperature=12026.05 | 36 | 70 | 58 | 36 | 42 | 84 | 48 | 56 | 26 | 42 | 49.8 |