---
pipeline_tag: robotics
---

# SparseDriveV2: Scoring is All You Need for End-to-End Autonomous Driving

SparseDriveV2 is an end-to-end multi-modal planning framework for autonomous driving. It demonstrates that performance consistently improves as trajectory anchors become denser, achieving state-of-the-art results through a scalable vocabulary representation and a factorized scoring strategy.

- **Paper:** [SparseDriveV2: Scoring is All You Need for End-to-End Autonomous Driving](https://huggingface.co/papers/2603.29163)
- **GitHub Repository:** [swc-17/SparseDriveV2](https://github.com/swc-17/SparseDriveV2)

## Method Overview

SparseDriveV2 pushes the performance boundary of scoring-based planning through two complementary innovations:
1.  **Scalable Vocabulary Representation:** A factorized structure that decomposes trajectories into geometric paths and velocity profiles, enabling combinatorial coverage of the action space.
2.  **Scalable Scoring Strategy:** A coarse factorized scoring over paths and velocity profiles followed by fine-grained scoring on a small set of composed trajectories.

This approach allows the model to scale its trajectory vocabulary to be 32x denser than prior methods while maintaining computational efficiency.

## Performance

The model achieves state-of-the-art performance using a lightweight ResNet-34 backbone:

| Benchmark | Metric | Score |
| :--- | :--- | :--- |
| **NAVSIM** | PDMS | 92.0 |
| **NAVSIM** | EPDMS | 90.1 |
| **Bench2Drive** | Driving Score | 89.15 |
| **Bench2Drive** | Success Rate | 70.00 |

## Citation

```bibtex
@article{sun2026sparsedrivev2,
  title={SparseDriveV2: Scoring is All You Need for End-to-End Autonomous Driving},
  author={Sun, Wenchao and Lin, Xuewu and Chen, Keyu and Pei, Zixiang and Li, Xiang and Shi, Yining and Zheng, Sifa},
  journal={arXiv preprint arXiv:2603.29163},
  year={2026}
}
```