Robotic Manipulation on libero object temp x0.1 (test)
0Execution Violation RateOracle abstention
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Oracle abstentionModel=OpenVLA-OFT, Seeds=5, Cells=2502026.05 | 0 | 91 | 83 | — | |
| BOKBOModel=OpenVLA-OFT, Epsilon (ε)=0.05, Seeds=5, Cells=2502026.05 | 3.1 | 80 | 71 | 86 | |
| Held-out threshold (no guarantee)Model=OpenVLA-OFT, Seeds=5, Cells=2502026.05 | 3.4 | 79 | 69 | — | |
| No abstention (K-sample only)Model=OpenVLA-OFT, Seeds=5, Cells=2502026.05 | 8.6 | 100 | 68 | — |