Research map / Continual learning / Continual fine-tuning of LLMs
Sentinel-VLA: A Metacognitive VLA Model with Active Status Monitoring for Dynamic Reasoning and Error Recovery
Wenhao Li, Xiu Su, Dan Niu, Yichao Cao, Hongyan Xu, Zhe Qu, Lei Fan, Shan You, Chang Xu
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 |
|---|---|---|---|
| add status monitor expert (ablation) | Sentinel-VLA · RLBench Seen | success rate: 60.7% -> 62.6% | Improved |
| train OpenVLA with EC-Gen error-recovery data (ablation) | OpenVLA · RLBench Seen | success rate: 35.6% -> 42.4% | Improved |
| add self-evolving continual learning with OC-Adapter (ablation) | Sentinel-VLA · Real-world | success rate: 54.0% -> 60.0% | Improved |
| add SECL with plain LoRA (no orthogonality) (ablation) | Sentinel-VLA · Real-world | success rate: 54.0% -> 44.7% | Hurt |
| add orthogonality constraint to online adapter (ablation) | SECL · Real-world | success rate: 44.7% -> 60.0% | Improved |
| apply Sentinel-VLA vs PI0 base | PI0 · RLBench Seen | success rate: 57.8% -> 63.5% | Improved |
| add random disturbances during evaluation | OpenVLA · RLBench | success rate: 35.6% -> 25.6% | Hurt |
| add random disturbances during evaluation | Sentinel-VLA · RLBench | success rate: 63.5% -> 54.7% | Hurt |
| use Sentinel-VLA instead of PI0 | PI0 · LIBERO-LONG | success rate: 85.2% -> 90.7% | Improved |
| use Sentinel-VLA instead of PI0 | PI0 · Real-world | success rate: 46.0% -> 60.0% | Improved |
| on-demand reasoning instead of chain-of-thought every step | ECoT · Real-world RTX4090 | computation time per action: 1528 ms -> 13 ms | Improved |
| add status monitor expert | Sentinel-VLA · Status monitor F1 (simulation vs real) | F1: 0.9024 -> 0.8567 | Mixed |
Closest papers by meaning
- VLA-ATTC: Adaptive Test-Time Compute for VLA Models with Relative Action Critic Model (ICML 2026 · similarity 0.90)
- VLA-Cache: Efficient Vision-Language-Action Manipulation via Adaptive Token Caching (NeurIPS 2025 · similarity 0.90)
- ElegantVLA: Learning When to Think for Efficient Vision-Language-Action Models (NeurIPS 2026 · similarity 0.89)
- UP-VLA: A Unified Understanding and Prediction Model for Embodied Agent (ICML 2025 · similarity 0.89)
- VLA-Corrector: Lightweight Detect-and-Correct Inference for Adaptive Action Horizon (NeurIPS 2026 · similarity 0.89)
- VLA-REPLICA: A Low-Cost, Reproducible Benchmark for Real-World Evaluation of Vision-Language-Action Models (NeurIPS 2026 · similarity 0.89)
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.
