Physical task routing on Physical Suite Intelligence/Thinking axis (test)
63.04Success Rate (Q)accuracy oracle
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| accuracy oracleCategory=Oracle2026.06 | 63.04 | 4.62 | 97.84 | |
| Trained RouterArchitecture=KNN / PR-KNN (k=5), Head=regression2026.06 | 60.87 | 8.78 | 86.34 | |
| Trained RouterArchitecture=Linear, Head=regression2026.06 | 58.7 | 9.8 | 77.73 | |
| Trained RouterArchitecture=OVR, Head=rank regression2026.06 | 54.35 | 6.9 | 64 | |
| Trained RouterArchitecture=MLP [2048,1024], Head=tie regression2026.06 | 52.17 | 8.23 | 55.04 | |
| Trained RouterArchitecture=K-means, Head=classification2026.06 | 52.17 | 10.86 | 53.64 | |
| Think plannerCategory=Fixed Baseline2026.06 | 52.17 | 19.19 | 0 | |
| OOD baselineCategory=OOD Baseline2026.06 | 50 | 17.65 | 32.57 | |
| randomCategory=Fixed Baseline2026.06 | 43.48 | 11.4 | 19.19 | |
| No-Think plannerCategory=Fixed Baseline2026.06 | 39.13 | 1.07 | 0 |