---
license: apache-2.0
language:
- en
tags:
- autonomous-driving
- end-to-end-driving
- diffusion-model
- trajectory-planning
- uncertainty-estimation
- navsim
pretty_name: Mimir Data
size_categories:
- 100M<n<1G
---

# Mimir: Hierarchical Goal-Driven Diffusion with Uncertainty Propagation for End-to-End Autonomous Driving

> [**Mimir: Hierarchical Goal-Driven Diffusion with Uncertainty Propagation for End-to-End Autonomous Driving**](https://arxiv.org/pdf/2512.07130)
>
> Zebin Xing, Yupeng Zheng, Qichao Zhang, Zhixing Ding, Pengxuan Yang, Songen Gu, Zhongpu Xia, Dongbin Zhao
>
> Institute of Automation, Chinese Academy of Sciences; China University of Geosciences; Fudan University
>
> IEEE Robotics and Automation Letters (RAL), 2025

<div align="center">
    <img src="./assets/teaser.png" width="75%" />
</div>

**Mimir** is a hierarchical goal-driven diffusion model for end-to-end autonomous driving. It improves upon **GoalFlow** by explicitly modeling goal uncertainty and accelerating goal inference, enabling more robust and efficient planning. With a ResNet-34 backbone, Mimir achieves 89.3 PDMS and 34.6 EPDMS, demonstrating strong performance in both accuracy and efficiency.



This repository contains the **model weights and intermediate data files** required for Mimir's evaluation and training. For the full codebase, please visit the [main Mimir repository](https://github.com/ZebinX/Mimir-Uncertainty-Driving).

---

## 📦 Repository Structure

| Path | Description |
|:---|:---|
| `ckpts/mimir_epoch94.ckpt` | Mimir planning model (~746MB) |
| `ckpts/mimir_unc_epoch99.ckpt` | Uncertainty modeling model (~724MB) |
| `goalflow_goal/` | GoalFlow goal points for `mimir_unc` training initialization (navhard/navtest/navtrain) |
| `goalflow_dac/` | Predicted DAC using GoalFlow (navhard/navtest/navtrain) |
| `navhard_naviunc/` | Navigation & uncertainty data for Navhard |
| `navstest_naviunc/` | Navigation & uncertainty data for Navtest |
| `navtrain_naviunc/` | Navigation & uncertainty data for Navtrain |
| `cluster_points_8192_.npy` | 8192 cluster centers for goal point discretization |
| `frame_mapping_navhard.yaml` | Frame index mapping for Navhard |
| `global_pose_navhard.npy` | Global pose data for Navhard |

## 📊 Results

<div align="center">
   <img src="./assets/navhard.png" width="49%" />
    <img src="./assets/navtest.png" width="49%" />

</div>

Please refer to the [paper](https://arxiv.org/pdf/2512.07130) for more details.

---

## 📄 Citation

If you find Mimir useful, please consider citing our paper:

```bibtex
@ARTICLE{11282450,
  author={Xing, Zebin and Zheng, Yupeng and Zhang, Qichao and Ding, Zhixing and Yang, Pengxuan and Gu, Songen and Xia, Zhongpu and Zhao, Dongbin},
  journal={IEEE Robotics and Automation Letters}, 
  title={Mimir: Hierarchical Goal-Driven Diffusion With Uncertainty Propagation for End-to-End Autonomous Driving}, 
  year={2026},
  volume={11},
  number={2},
  pages={2178-2185},
  keywords={Uncertainty;Trajectory;Predictive models;Autonomous vehicles;Laser radar;Vocabulary;Planning;Feature extraction;Estimation;Artificial intelligence;Learning from demonstration;imitation learning;autonomous vehicle navigation},
  doi={10.1109/LRA.2025.3641129}}
```

## Contact

If you have any questions or suggestions, please open an issue or contact: xzebin@bupt.edu.cn