LIBERO-PeRM Zero-shot - All-Pref SA: leaderboard
LIBERO-PeRM v1 Table2; zero-shot LIBERO-SinglePref robot-action policies on edited LIBERO layouts; No-Pref uses original task language. SA = mean(task-success indicator * discrete preference label 0,0.25,0.75,1), not conditional success-only adherence. Metric: Success-gated preference satisfaction (%). Higher is better. Source: https://arxiv.org/html/2610.22372v1
Metric: Success-gated preference satisfaction (%). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | pi0.5 | 37.92 | |
| 2 | pi0-fast | 30.37 | |
| 3 | GR00T-N1.6 | 27.45 | |
| 4 | UniVLA | 21.52 | |
| 5 | OpenVLA-OFT | 4.53 | |
| 6 | SmolVLA | 2.03 |
Interactive version: theaggregate.ai/benchmark?slug=libero-perm-zero-shot-all-pref-sa · How It Works · Data refreshed daily, snapshot 2026-10-17.