---
library_name: transformers
pipeline_tag: image-text-to-text
base_model: Qwen/Qwen3-4B-Instruct-2507
---

# EMBGuard

EMBGuard is the first MLLM-based safety guardrail for embodied agents designed to decouple physical risk reasoning from agent policy. By evaluating a (visual observation, action) pair, EMBGuard identifies hazardous configurations and provides natural language explanations of potential risks.

- **Paper:** [EMBGuard: Constructing Hazard-Aware Guardrails for Safe Planning in Embodied Agents](https://huggingface.co/papers/2605.30924)
- **Repository:** [https://github.com/dongwxxkchoi/EMBGuard](https://github.com/dongwxxkchoi/EMBGuard)

## Model Description
MLLM-powered embodied agents deployed in real-world environments encounter physical hazards. However, existing approaches lack explicit mechanisms for identifying hazards and reasoning about action-conditioned risks, leading agents to either miss risky interactions or over-identify risks. 

EMBGuard identifies hazardous configurations across seven physical risk categories and provides natural language explanations of potential risks. Despite its compact size, EMBGuard achieves performance competitive with proprietary MLLMs (e.g., GPT-5.1, Gemini-2.5-Pro) while significantly reducing false-positive rates, making it suitable for real-time deployment in safety-critical planning.

The model was trained on **EMBHazard**, a dataset of 15.1K action-conditioned pairs, and evaluated on **EMBGuardTest**, a benchmark of 329 manually curated real-world scenarios.

## Usage
For detailed instructions on installation, training, and evaluation, please refer to the [official GitHub repository](https://github.com/dongwxxkchoi/EMBGuard).

## Citation
If you use EMBGuard in your research, please cite:

```bibtex
@article{choi2024embguard,
  title={EMBGuard: Constructing Hazard-Aware Guardrails for Safe Planning in Embodied Agents},
  author={Choi, Dongwook and others},
  journal={arXiv preprint arXiv:2605.30924},
  year={2024}
}
```