README.md

---
license: apache-2.0
language:
- en
library_name: pytorch
model_name: Copper-Policy
tags:
- robotics
- robot-manipulation
- world-action-model
- libero
- libero-plus
- robotwin
- arxiv:2609.32779
---

<h1 align="center"><img src="assets/copper.png" width="32" height="32" align="absmiddle" style="display: inline-block; vertical-align: middle;" alt="Copper-Policy logo"> Copper-Policy</h1>

<h3 align="center">Focus on the Representation for Robust Robot Manipulation</h3>

<p align="center"><b>Zexin Feng · Yixu Feng · Lingyu Xiao · Shang Su · Kexin Zheng<br>Chang Xu · Mengkai Shi · Shuo Feng · Xintao Yan</b></p>

<p align="center">The University of Hong Kong · The University of Sydney · Tsinghua University · DenseAI</p>

<p align="center"><a href="https://huggingface.co/Mark455/Copper-Policy/blob/main/README.md">English</a> | <a href="https://huggingface.co/Mark455/Copper-Policy/blob/main/README_zh.md">中文</a></p>

Official inference checkpoints for **Copper-Policy: Focus on the Representation for Robust Robot Manipulation**.

Copper-Policy combines a compact world representation with spatial visual features for robot manipulation. This release contains two approximately 2B-parameter policies.

<p align="center"><a href="https://arxiv.org/abs/2609.32779">📄 Paper</a> · <a href="https://zexinfeng-cn.github.io/works/copper-policy/">🌐 Project Page</a> · <a href="https://github.com/Mark4551124015/Copper-Policy">💻 Code</a></p>

## 📦 Included checkpoints

```text
libero/
├── policy.pt
└── policy.json
robotwin/
├── policy.pt
└── policy.json
release_manifest.json
SHA256SUMS
LICENSE
```

- **LIBERO:** one checkpoint shared by LIBERO and LIBERO-Plus; 2 RGB views, 7-dimensional actions and 8-dimensional proprioception.
- **RoboTwin:** one checkpoint shared by clean and randomized evaluations; 3 RGB views, 14-dimensional actions and proprioception.
- Both checkpoints predict an action horizon of **32**. Normalization statistics and the configuration needed for inference are included in `policy.pt`; no separate dataset-statistics file is required.
- `policy.json` describes each checkpoint; `SHA256SUMS` provides checksums. These custom PyTorch checkpoints are loaded using the Copper-Policy code repository.

## 📥 Download

<div align="center">

| Checkpoint | Benchmarks |
| --- | --- |
| `libero/policy.pt` | LIBERO / LIBERO-Plus |
| `robotwin/policy.pt` | RoboTwin clean / randomized |

</div>

Each checkpoint is approximately **4.06 GB**. Both together require approximately **8.12 GB**.

```bash
# Run in the Copper-Policy code repository
uv run --frozen --all-extras --no-sync hf download Mark455/Copper-Policy \
  libero/policy.pt robotwin/policy.pt \
  --local-dir pretrained_weights/copper_policy
```

To download only one model, keep only its file path in the command.

For installation and inference, see the [code repository](https://github.com/Mark4551124015/Copper-Policy).

## ⚙️ Inference

The code repository provides four evaluation launchers:

```bash
uv run --frozen --all-extras --no-sync bash test_libero.sh --all-tasks
uv run --frozen --all-extras --no-sync bash test_libero_plus.sh --all-tasks
uv run --frozen --all-extras --no-sync bash test_robotwin_clean.sh --all-tasks
uv run --frozen --all-extras --no-sync bash test_robotwin_rand.sh --all-tasks
```

Default inference uses **10 denoising steps**, with **10 replan steps for LIBERO / LIBERO-Plus** and **24 for RoboTwin**. Compilation and rollout videos are enabled. Download the V-JEPA 2.1, DINOv2 and Wan T5 encoder presets using `python -m tools.download_weights all --yes` through the repository's uv environment. No separate Wan transformer initialization weights are needed.

Follow the [code repository](https://github.com/Mark4551124015/Copper-Policy) for installation and simulator assets; adapt PyTorch/CUDA to your hardware and driver. **Training code — coming soon.**

<details>
<summary><b>Note: LIBERO language mapping</b></summary>

LIBERO training instructions sometimes differ from benchmark prompts. The code maps exact plaintext benchmark prompts to training instructions. LIBERO-Plus language perturbations are not mapped; unmatched prompts retain the benchmark text.

</details>

## 📊 Reported results

<div align="center">

| LIBERO | LIBERO-Plus | RoboTwin clean | RoboTwin randomized |
| :---: | :---: | :---: | :---: |
| 97.25% | 80.85% | 70.84% | 12.98% |

</div>

These are results reported in the [paper and project page](https://zexinfeng-cn.github.io/works/copper-policy/). Individual evaluation runs record their own measured scores.

## 📚 Citation

```bibtex
@article{feng2026copper,
  title={Copper-Policy: Focus on the Representation for Robust Robot Manipulation},
  author={Feng, Zexin and Feng, Yixu and Xiao, Lingyu and Su, Shang and Zheng, Kexin and Xu, Chang and Shi, Mengkai and Feng, Shuo and Yan, Xintao},
  journal={arXiv preprint arXiv:2609.32779},
  year={2026}
}
```

## 🙏 Acknowledgements and license

Our implementation builds on [FastWAM](https://github.com/yuantianyuan01/FastWAM) and [LingBot-VA](https://github.com/robbyant/lingbot-va). Dataset conversion uses our [lerobot-tools](https://github.com/Mark4551124015/lerobot-tools). Codex helped organize the open-source code.

Released under [Apache 2.0](LICENSE). External encoders and simulator assets retain their respective licenses.