README.md

---
license: mit
size_categories:
- 1M<n<10M
pretty_name: LEAD Carla Leaderboard 2.0
task_categories:
- robotics
tags:
- autonomous-driving
- imitation-learning
- carla
- transfuser
---

# LEAD: Minimizing Learner–Expert Asymmetry in End-to-End Driving

[**Project Page**](https://ln2697.github.io/lead) | [**Paper**](https://huggingface.co/papers/2512.20563) | [**Code**](https://github.com/autonomousvision/lead)

Official CARLA dataset accompanies our paper LEAD: Minimizing Learner–Expert Asymmetry in End-to-End Driving.

> We release the complete pipeline required to achieve state-of-the-art closed-loop performance on the Bench2Drive benchmark. Built around the CARLA simulator, the stack features a data-centric design with:
> 
> - Extensive visualization suite and runtime type validation for easier debugging.
> - Optimized storage format, packs 72 hours of driving in ~200GB.
> - Native support for NAVSIM and Waymo Vision-based E2E and extending those benchmarks through closed-loop simulation and synthetic data for additional supervision during training.

Find more information on [https://github.com/autonomousvision/lead](https://github.com/autonomousvision/lead).

## Format

Each route is stored as a sequence of synchronized frames. All sensor modalities are ego-centric and time-aligned.
In addition to the nominal sensor suite, we provide a second, perturbated sensor stack corresponding to a counterfactual ego state used for recovery supervision.

```html
├── bboxes/                  # Per-frame 3D bounding boxes for all actors
├── depth/                   # Compressed depth maps (should be used for auxiliary supervision only)
├── depth_perturbated        # Depth from a perturbated ego state
├── hdmap/                   # Ego-centric rasterized HD map
├── hdmap_perturbated        # HD map aligned to perturbated ego pose
├── lidar/                   # LiDAR point clouds
├── metas/                   # Per-frame metadata and ego state
├── radar/                   # Radar detections
├── radar_perturbated        # Radar detections from perturbated ego state
├── rgb/                     # Front-facing RGB images
├── rgb_perturbated          # RGB images from perturbated ego state
├── semantics/               # Semantic segmentation maps
├── semantics_perturbated    # Semantics from perturbated ego state
└── results.json             # Route-level summary and evaluation metadata
```

## Download

You can either download a **single route** (useful for quick inspection / debugging) or **clone the full dataset** via Git LFS and unzip all routes.

**Note:** Download the dataset after setting up the [lead repository](https://github.com/autonomousvision/lead).

### Option 1: Download a single route

```bash
bash scripts/download_one_route.sh
```

### Option 2: Download all routes (Git LFS)

Clone the dataset repository directly into the expected directory:

```bash
git lfs install
git clone https://huggingface.co/datasets/ln2697/lead_carla data/carla_leaderboard2/zip
```

### Unzip routes

Run

```bash
bash scripts/unzip_routes.sh
```

## Citation

If you find this work useful, please cite:

```bibtex
@inproceedings{Nguyen2026CVPR,
	author = {Long Nguyen and Micha Fauth and Bernhard Jaeger and Daniel Dauner and Maximilian Igl and Andreas Geiger and Kashyap Chitta},
	title = {LEAD: Minimizing Learner-Expert Asymmetry in End-to-End Driving},
	booktitle = {Conference on Computer Vision and Pattern Recognition (CVPR)},
	year = {2026},
}
```

## License

This project is released under the [MIT License](LICENSE)