Research map / Vision-language and multimodal / Robot action and navigation
VLA Knows Its Limits
Haoxuan Wang, Gengyu Zhang, Yan Yan, Ramana Kompella, Gaowen Liu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace random horizon with attention-guided AutoHorizon (ablation) | pi0.5 (p=50) · LIBERO-Spatial (p=50) | success rate (%): 82.8 -> 96.5 | Improved |
| replace random horizon with attention-guided AutoHorizon (ablation) | pi0.5 (p=10) · LIBERO-10 (p=10) | success rate (%): 86.3 -> 91.6 | Improved |
| lengthen fixed execution horizon from 0.2p to 1.0p (ablation) | pi0.5 (p=50) · LIBERO-Spatial (p=50) | success rate (%): 94.9 -> 71.5 | Hurt |
| lengthen fixed execution horizon from 0.2p to 1.0p (ablation) | pi0.5 (p=10) · LIBERO-10 (p=10) | success rate (%): 76.0 -> 90.4 | Improved |
| use AutoHorizon instead of brute-force Static Oracle+ (ablation) | pi0.5 (p=50) · LIBERO-10 (p=50) | success rate (%): 91.9 -> 92.1 | No clear effect |
| use AutoHorizon instead of brute-force Static Oracle+ (ablation) | GR00T N1.5 · LIBERO-Goal | success rate (%): 94.7 -> 96.0 | Improved |
| use AutoHorizon instead of Static Oracle+ (ablation) | GR00T N1.5 · LIBERO-10 | success rate (%): 90.0 -> 92.7 | Improved |
| use AutoHorizon instead of Static Oracle+ (ablation) | pi0.5 · RoboTwin Adjust Bottle | success rate (%): 98.7 -> 100.0 | Improved |
| use AutoHorizon instead of Static Oracle+ (ablation) | pi0.5 (real robot) · Real-world Cube to Plate | stage-based solve rate (%): 81.5 -> 92.0 | Improved |
| very short fixed execution horizon e=5 (ablation) | pi0.5 (real robot) · Real-world Cube to Plate | stage-based solve rate (%): 81.5 -> 0.0 | Hurt |
| random horizon vs best fixed horizon (ablation) | pi0.5 (p=50) · LIBERO-Goal | success rate (%): 94.4 -> 85.3 | Hurt |
| use AutoHorizon vs Random on real robot (ablation) | pi0.5 (real robot) · Real-world Cube to Bowl | stage-based solve rate (%): 77.0 -> 99.0 | Improved |
Closest papers by meaning
- FASTER: Rethinking Real-Time Flow VLAs (NeurIPS 2026 · similarity 0.92)
- Not All Features Are Created Equal: A Mechanistic Study of Vision-Language-Action Models (NeurIPS 2026 · similarity 0.92)
- VLA-Corrector: Lightweight Detect-and-Correct Inference for Adaptive Action Horizon (NeurIPS 2026 · similarity 0.91)
- ElegantVLA: Learning When to Think for Efficient Vision-Language-Action Models (NeurIPS 2026 · similarity 0.91)
- VLA-ATTC: Adaptive Test-Time Compute for VLA Models with Relative Action Critic Model (ICML 2026 · similarity 0.91)
- Real-Time Execution of Action Chunking Flow Policies (NeurIPS 2025 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
