Add model card and robotics metadata

README.md

@@ -0,0 +1,54 @@
+---
+pipeline_tag: robotics
+library_name: lerobot
+tags:
+- vla
+- smolvla
+- vgas
+- vision-language-action
+---
+
+# VGAS: Value-Guided Action-Chunk Selection
+
+This repository contains a Vision-Language-Action (VLA) policy based on the **VGAS** framework, as presented in the paper [VGAS: Value-Guided Action-Chunk Selection for Few-Shot Vision-Language-Action Adaptation](https://huggingface.co/papers/2602.07399).
+
+VGAS (Value-Guided Action-chunk Selection) is a framework designed for few-shot VLA adaptation. It uses a generation-selection approach, employing a finetuned VLA as a high-recall proposal generator and a **Q-Chunk-Former** critic to identify action chunks that are both semantically faithful and geometrically precise during inference.
+
+## Resources
+- **Paper:** [VGAS: Value-Guided Action-Chunk Selection for Few-Shot Vision-Language-Action Adaptation](https://huggingface.co/papers/2602.07399)
+- **Code:** [GitHub Repository](https://github.com/Jyugo-15/VGAS)
+
+## Usage
+
+This model is compatible with the [LeRobot](https://github.com/huggingface/lerobot) ecosystem. To evaluate the policy using the scripts provided in the official repository:
+
+### Evaluation (Behavior Cloning - No Critic)
+```bash
+python scripts/eval_qc_bestofn.py \
+  --env-task libero_goal \
+  --policy-path SemyonXu616/my_smolvla \
+  --no-use-best-of-n \
+  --best-of-n 1
+```
+
+### Evaluation (Best-of-N - Critic Guided)
+To perform inference-time selection as described in the paper, you will also need the critic checkpoint:
+```bash
+python scripts/eval_qc_bestofn.py \
+  --env-task libero_goal \
+  --policy-path SemyonXu616/my_smolvla \
+  --critic-state <CRITIC_CKPT> \
+  --use-best-of-n \
+  --best-of-n 8
+```
+
+## Citation
+If you find this work useful, please cite:
+```bibtex
+@article{vgas2025,
+  title={VGAS: Value-Guided Action-Chunk Selection for Few-Shot Vision-Language-Action Adaptation},
+  author={Jiuhai Chen and Jianwei Yang and Haiping Wu and Dianqi Li and Jianfeng Gao and Tianyi Zhou and Bin Xiao},
+  journal={arXiv preprint arXiv:2602.07399},
+  year={2025}
+}
+```