README.md

---
pipeline_tag: robotics
license: apache-2.0
tags:
- reinforcement-learning
- robotic-manipulation
- action-chunking
---

# Mixture of Horizons in Action Chunking

This repository hosts the official implementation of **Mixture of Horizons (MoH)**, introduced in the paper [Mixture of Horizons in Action Chunking](https://huggingface.co/papers/2511.19433).

Vision-language-action (VLA) models for robotic manipulation are highly sensitive to the chosen **action chunk length**, termed **horizon** in this work. A fixed horizon presents an inherent trade-off: longer horizons offer superior global foresight but compromise fine-grained accuracy, while shorter ones provide precise local control but struggle with long-term tasks.

To address this challenge, we propose **Mixture of Horizons (MoH)**, a novel, plug-and-play strategy that fuses multiple horizons within a single policy. MoH processes action chunks in parallel segments with different horizons and integrates their outputs. This approach simultaneously leverages long-term foresight and short-term precision with minimal overhead, and enables **Dynamic Inference** through cross-horizon consensus for enhanced efficiency and robustness in complex robotic tasks.

-   📄 [Paper](https://huggingface.co/papers/2511.19433)
-   📝 [Project Page](https://timsty1.github.io/moh/)
-   💻 [Code](https://github.com/Timsty1/MixtureOfHorizons/tree/main)

## Introduction

<div align="center">
  <table border="0" cellspacing="0" cellpadding="0">
    <tr>
      <td align="center" width="50%">
        <img src="https://github.com/Timsty1/MixtureOfHorizons/raw/main/figure/study_of_horizons_pi0.png" alt="Trade-off Effect" width="100%">
      </td>
      <td align="center" width="50%">
        <img src="https://github.com/Timsty1/MixtureOfHorizons/raw/main/figure/intro_motivation_v2.png" alt="Mixture of Horizons" width="100%">
      </td>
    </tr>
    <tr>
      <td align="center" valign="top">
        Figure 1: Trade-off between long-term foresight and short-term precision induced by single horizon
      </td>
      <td align="center" valign="top">
        Figure 2: Overview of the proposed mixture-of-horizons strategy
      </td>
    </tr>
  </table>
</div>

<br>

*   **Mitigates Trade-off**: Addresses the inherent trade-off between long-term foresight and short-term precision induced by single action chunk horizons.
*   **Plug-and-Play**: Easily integrates into existing full-attention action modules with minimal training or inference overhead.
*   **Dynamic Inference**: Achieves higher efficiency and robustness by selecting stable actions through cross-horizon consensus.

#### More results on LIBERO
<div align="center">
<img src="https://github.com/Timsty1/MixtureOfHorizons/raw/main/figure/libero_main.jpg" width="90%" />
</div>

## Usage

For detailed instructions on environment setup, training, and evaluation, please refer to the [GitHub repository](https://github.com/Timsty1/MixtureOfHorizons/tree/main).

## ❤️ Acknowledgment

We express our gratitude to [OpenPi](https://github.com/Physical-Intelligence/openpi/tree/main), [LIBERO](https://github.com/Lifelong-Robot-Learning/LIBERO), and [RoboTwin](https://robotwin-platform.github.io/) for their open-source contributions.

## 📝 Citation

If you feel that this paper, models, or codes are helpful, please cite our paper, thanks for your support!

```bibtex
@article{jing2025mixture_of_horizons,
  title={Mixture of Horizons in Action Chunking},
  author={Jing, Dong and Wang, Gang and Liu, Jiaqi and Tang, Weiliang and Sun, Zelong and Yao, Yunchao and Wei, Zhenyu and Liu, Yunhui and Lu, Zhiwu and Ding, Mingyu},
  journal={arXiv preprint arXiv:2511.19433},
  year={2025}
}
```