---
license: mit
---
# Epona: Autoregressive Diffusion World Model for Autonomous Driving
This repository contains the model checkpoints and test meta data in the paper [Epona: Autoregressive Diffusion World Model for Autonomous Driving
](https://arxiv.org/abs/2506.24113.pdf).

🌏 Project page : [https://kevin-thu.github.io/Epona/](https://kevin-thu.github.io/Epona/).

⭐ Code: [https://github.com/Kevin-thu/Epona](https://github.com/Kevin-thu/Epona).

It contains four model checkpoints:
- `epona_nuplan.pkl`: The base world model trained on NuPlan from scratch;
- `epona_nuplan+nusc.pkl`: The world model finetuned on NuScenes based on the base model;
- `epona_nuplan+china.pkl`: The world model fintuned on in-house China driving data based on the base model;
- `dcae_td_20000.pkl`: The temporal-aware DCAE (downsampling rate=32, channel dim=32, temporal patch size=6) fintuned on NuPlan video clips.  

And two folders of preprocessed json files:
- `test_meta_data_nuplan`: Preprocessed meta infos on NuPlan test set (not including train set);
- `meta_data_nusc`: Preprocessed meta infos on NuScenes train set and validation set.