# SimScale **Repository Path**: tj1652045/SimScale ## Basic Information - **Project Name**: SimScale - **Description**: No description available - **Primary Language**: Unknown - **License**: Apache-2.0 - **Default Branch**: main - **Homepage**: None - **GVP Project**: No ## Statistics - **Stars**: 0 - **Forks**: 0 - **Created**: 2026-04-21 - **Last Updated**: 2026-04-21 ## Categories & Tags **Categories**: Uncategorized **Tags**: None ## README

# **Learning to Drive via Real-World Simulation at Scale** [![Paper](https://img.shields.io/badge/ArXiv-A42C25?style=for-the-badge&logo=arxiv&logoColor=white)](https://arxiv.org/abs/2511.23369) [![Home](https://img.shields.io/badge/project_page-5F259F?style=for-the-badge&logo=homepage&logoColor=white)](https://opendrivelab.com/SimScale/) [![Hugging Face](https://img.shields.io/badge/hugging_face-ffc107?style=for-the-badge&logo=huggingface&logoColor=white)](https://huggingface.co/datasets/OpenDriveLab/SimScale) [![ModelScope](https://img.shields.io/badge/modelscope-624AFF?style=for-the-badge&logo=modelscope&logoColor=white)](https://modelscope.cn/datasets/OpenDriveLab/SimScale) [![License](https://img.shields.io/badge/Apache--2.0-019B8F?style=for-the-badge&logo=apache)](https://github.com/OpenDriveLab/SimScale/blob/main/LICENSE)

> [Haochen Tian](https://github.com/hctian713), > [Tianyu Li](https://sephyli.github.io/), > [Haochen Liu](https://georgeliu233.github.io/), > [Jiazhi Yang](https://github.com/YTEP-ZHI), > [Yihang Qiu](https://github.com/gihharwtw), > [Guang Li](https://scholar.google.com/citations?user=McEfO8UAAAAJ&hl=en), > [Junli Wang](https://openreview.net/profile?id=%7EJunli_Wang4), > [Yinfeng Gao](https://scholar.google.com/citations?user=VTn0hqIAAAAJ&hl=en), > [Zhang Zhang](https://scholar.google.com/citations?user=rnRNwEMAAAAJ&hl=en), > [Liang Wang](https://scholar.google.com/citations?user=8kzzUboAAAAJ&hl=en), > [Hangjun Ye](https://scholar.google.com/citations?user=68tXhe8AAAAJ&hl=en), > [Tieniu Tan](https://scholar.google.com/citations?user=W-FGd_UAAAAJ&hl=en), > [Long Chen](https://long.ooo/), > [Hongyang Li](https://lihongyang.info/) > > > - πŸ“§ Primary Contact: Haochen Tian (tianhaochen2023@ia.ac.cn) > - πŸ“œ Materials: 🌐 [𝕏](https://x.com/OpenDriveLab/status/1999507869633527845) | πŸ“° [Media](https://mp.weixin.qq.com/s/OGV3Xlb0bHSSSloG11qFJA) | πŸ—‚οΈ [Slides](https://docs.google.com/presentation/d/17qbsKZU9jdw7MfiPk7hZelaLb3leR2M76gPcMkuf1MI/edit?usp=sharing) | πŸͺ§ [Poster](https://docs.google.com/presentation/d/1OrEj_llLyHPK8uSj_tmiam5T3BMYXoNPUjEw5h1_slk/edit?usp=sharing) | 🎬 [Talk (in Chinese)](https://www.bilibili.com/video/BV1tqrEBNECQ) > - πŸ–ŠοΈ Joint effort by CASIA, OpenDriveLab at HKU, and Xiaomi EV. --- ## πŸ”₯ Highlights - πŸ—οΈ A scalable simulation pipepline that synthesizes diverse and high-fidelity reactive driving scenarios with pseudo-expert demonstrations. - πŸš€ An effective sim-real co-training strategy that improves robustness and generalization synergistically across various end-to-end planners. - πŸ”¬ A comprehensive recipe that reveals crucial insights into the underlying scaling properties of sim-real learning systems for end-to-end autonomy. ## πŸ“’ News - **`[2026/4/9]`** πŸŽ‰πŸŽ‰πŸŽ‰ Awarded as CVPR 2026 Oral. - **`[2026/2/21]`** πŸŽ‰ Accepted to CVPR 2026. - **`[2026/1/16]`** We released the data and models on πŸ‘Ύ ModelScope to better serve users in China. - **`[2026/1/6]`** We released the code **v1.0**. - **`[2025/12/31]`** We released the data and models **v1.0** on πŸ€— Hugging Face. Happy New Year ! πŸŽ„ - **`[2025/12/1]`** We released our [paper](https://arxiv.org/abs/2511.23369) on arXiv. ## πŸ“‹ TODO List - [x] More Visualization Results. - [x] Future Sensors Data. - [x] Sim-Real Co-training Code release (Jan. 2026). - [x] Simulation Data release (Dec. 2025). - [x] Checkpoints release (Dec. 2025). --- ## πŸ“Œ Table of Contents - πŸ›οΈ [Model Zoo](#%EF%B8%8F-model-zoo) - 🎯 [Getting Started](#-getting-started) - πŸ“¦ [Data Preparation](#-data-preparation) - [Download Dataset](#1-download-dataset) - [Set Up Configuration](#2-set-up-configuration) - βš™οΈ [Sim-Real Co-Training](#%EF%B8%8F-sim-real-co-training-recipe) - [Co-Training with Pseudo-Expert](#co-training-with-pseudo-expert) - [Co-Training with Rewards Only](#co-training-with-rewards-only) - πŸ” [Inference](#-inference) - [NAVSIM v2 navhard](#navsim-v2-navhard) - [NAVSIM v2 navtest](#navsim-v2-navtest) - ⭐ [License and Citation](#-license-and-citation) ## πŸ›οΈ Model Zoo
Model Backbone Sim-Real Config NAVSIM v2 navhard NAVSIM v2 navtest
EPDMS CKPT EPDMS CKPT
LTF ResNet34 w/ pseudo-expert 30.3 | +6.9 HF / MS 84.4 | +2.9 HF / MS
DiffusionDrive ResNet34 w/ pseudo-expert 32.6 | +5.1 HF / MS 85.9 | +1.7 HF / MS
GTRS-Dense ResNet34 w/ pseudo-expert 46.1 | +7.8 HF / MS 84.0 | +1.7 HF / MS
rewards only 46.9 | +8.6 HF / MS 84.6 | +2.3 HF / MS
V2-99 w/ pseudo-expert 47.7 | +5.8 HF / MS 84.5 | +0.5 HF / MS
rewards only 48.0 | +6.1 HF / MS 84.8 | +0.8 HF / MS
> [!NOTE] > We fixed a minor error in the simulation process without changing the method, resulting in better performance than the numbers reported in the early arXiv version v2. We have updated the arXiv version v2. ## 🎯 Getting Started ### 1. Clone SimScale Repo ```bash git clone https://github.com/OpenDriveLab/SimScale.git cd SimScale ``` ### 2. Create Environment ```bash conda env create --name simscale -f environment.yml conda activate simscale pip install -e . ``` ## πŸ“¦ Data Preparation Our released simulation data is based on [nuPlan](https://www.nuscenes.org/nuplan) and [NAVSIM](https://github.com/autonomousvision/navsim). **We recommend first preparing the real-world data by following the instructions in [Download NAVSIM](https://github.com/autonomousvision/navsim/blob/main/docs/install.md#2-download-the-dataset). If you plan to use GTRS, please directly refer [Download NAVSIM](./docs/install.md#2-download-the-dataset).** ### 1. Download Dataset We provide πŸ€— [Script (Hugging Face)](./tools/download_hf.sh) and πŸ‘Ύ [Script (ModelScope)](./tools/download_ms) (users in China) for downloading the simulation data . Our simulation data format follows that of [OpenScene](https://github.com/OpenDriveLab/OpenScene/blob/main/docs/getting_started.md#download-data), with each clip/log has a fixed temporal horizon of 6 seconds at 2 Hz (2 s history + 4 s future), which are stored separately in `sensor_blobs_hist` and `sensor_blobs_fut`, respectively. **For policy training, `sensor_blobs_hist` alone is sufficient.** #### πŸ“Š Overview Table of Simulated Synthetic Data
Split / Sim. Round # Tokens Logs Sensors_Hist Sensors_Fut Link
Planner-based Pseudo-Expert
reaction_pdm_v1.0-0 65K 9.9GB 569GB 1.2T HF+ HF_Fut / MS
reaction_pdm_v1.0-1 55K 8.5GB 448GB 964GB HF+ HF_Fut / MS
reaction_pdm_v1.0-2 46K 6.9GB 402GB 801GB HF+ HF_Fut / MS
reaction_pdm_v1.0-3 38K 5.6GB 333GB 663GB HF+ HF_Fut / MS
reaction_pdm_v1.0-4 32K 4.7GB 279GB 554GB HF+ HF_Fut / MS
Recovery-based Pseudo-Expert
reaction_recovery_v1.0-0 45K 6.8GB 395GB 789GB HF+ HF_Fut / MS
reaction_recovery_v1.0-1 36K 5.5GB 316GB 631GB HF+ HF_Fut / MS
reaction_recovery_v1.0-2 28K 4.3GB 244GB 488GB HF+ HF_Fut / MS
reaction_recovery_v1.0-3 22K 3.3GB 189GB 378GB HF+ HF_Fut / MS
reaction_recovery_v1.0-4 17K 2.7GB 148GB 296GB HF+ HF_Fut / MS
> [!TIP] > Before downloading, we recommend checking the table above to select the appropriate split and `sensor_blobs`. #### 🏭 Simulation Data Pipeline

#### 🧩 Examples of Simulated Synthetic Data

5c9694f15f9c5537

367cfa28901257ee

d37c49db3dcd59fa

Sim. 1


Sim. 2


Sim. 3

Sim. 1


Sim. 2


Sim. 3

Sim. 1


Sim. 2


Sim. 3
### 2. Set Up Configuration We provide a [Script](./tools/move.sh) for moving the download simulation data to create the following structure. ```angular2html navsim_workspace/ β”œβ”€β”€ simscale/ β”œβ”€β”€ exp/ └── dataset/ β”œβ”€β”€ maps/ β”œβ”€β”€ navsim_logs/ β”‚ β”œβ”€β”€ test/ β”‚ β”œβ”€β”€ trainval/ β”‚ β”œβ”€β”€ synthetic_reaction_pdm_v1.0-*/ β”‚ β”‚ β”œβ”€β”€ [log]-00*.pkl β”‚ β”‚ └── ... β”‚ └── synthetic_reaction_recovery_v1.0-*/ β”œβ”€β”€ sensor_blobs/ β”‚ β”œβ”€β”€ test/ β”‚ β”œβ”€β”€ trainval/ β”‚ β”œβ”€β”€ synthetic_reaction_pdm_v1.0-*/ β”‚ β”‚ └── [token]-00*/ β”‚ β”‚ β”œβ”€β”€ CAM_B0/ β”‚ β”‚ └── ... β”‚ └── synthetic_reaction_recovery_v1.0-*/ └── navhard_two_stage/ ``` ## βš™οΈ Sim-Real Co-Training Recipe ### Preparation 1. Refer the [Script](./scripts/training/run_dataset_cache.sh) to cache the real-world and simulation data. 2. Download pretrained image backbone weight, [ResNet34](https://huggingface.co/timm/resnet34.a1_in1k) or [V2_99](https://drive.google.com/file/d/1gQkhWERCzAosBwG5bh2BKkt1k0TJZt-A/view). ### Co-Training with Pseudo-Expert We provide [Scripts](./scripts/training) for sim–real co-training, *e.g.*, [run_diffusiondrive_training_syn.sh](./scripts/training/run_diffusiondrive_training_syn.sh). The main configuration options are as follows: ```bash export SYN_IDX=0 # 0, 1, 2, 3, 4 export SYN_GT=pdm # pdm, recovery ``` - `SYN_IDX` specifies which rounds of simulation data are included; *e.g.*, `SYN_IDX=2` means that rounds 0, 1, and 2 will be used. - `SYN_GT` specifies the type of pseudo-expert used for supervision. In addition, the cache path for simulation data is hard-coded in [dataset.py#136](./navsim/planning/training/dataset.py#L136). Please make sure the path is correctly set to your local simulation data directory before training. - **Regression-based Policy | *LTF*** We provide a [Script](./scripts/training/run_transfuser_training_syn.sh) to train LTF with 8 GPUs for 100 epochs. - **Diffusion-based Policy | *DiffusionDrive*** We provide a [Script](./scripts/training/run_diffusiondrive_training_syn.sh) to train DiffusionDrive with 8 GPUs for 100 epochs. - **Scoring-based Policy | *GTRS-Dense*** We provide a [Script](./scripts/training/run_gtrs_dense_training_multi_syn.sh) to train GTRS_Dense on 4 nodes, each with 8 GPUs, for 50 epochs. We also provide πŸ€— [Reward Files (Hugging Face)](https://huggingface.co/datasets/OpenDriveLab/SimScale/tree/main/SimScale_rewards) and πŸ‘Ύ [Reward Files (ModelScope)](https://www.modelscope.cn/datasets/OpenDriveLab/SimScale/tree/master/SimScale_rewards) (users in China) for rewards in simulation data. Please download correspending files first and move them to `NAVSIM_TRAJPDM_ROOT/sim`. The reward files path is hard-coded in [gtrs_agent.py#223](./navsim/agents/gtrs_dense/gtrs_agent.py#223). Check it before training. ### Co-Training with Rewards Only - **Scoring-based Policy | *GTRS-Dense*** It uses the same training [Script](./scripts/training/run_gtrs_dense_training_multi_syn.sh), to train GTRS_Dense on 4 nodes, each with 8 GPUs, for 50 epochs. The main configuration option is as follows: ```bash syn_imi=false # true, false ``` - `syn_imi`: When set to `false`, the imitation learning loss is disabled for simulation data, while it remains enabled for real-world data. ## πŸ” Inference ### Preparation Refer the [Script](./scripts/evaluation/run_metric_caching.sh) to cache metric first. ### NAVSIM v2 navhard We provide [Scripts](./scripts/evaluation_navhard) to evaluate three policies on [navhard](./navsim/planning/script/config/common/train_test_split/scene_filter/navhard_two_stage.yaml) using GPU inference. ### NAVSIM v2 navtest We provide [Scripts](./scripts/evaluation_navtest) to evaluate three policies on [navtest](./navsim/planning/script/config/common/train_test_split/scene_filter/navtest.yaml) using GPU inference. ## ❀️ Acknowledgements We acknowledge all the open-source contributors for the following projects to make this work possible: - [NAVSIM](https://github.com/autonomousvision/navsim) | [MTGS](https://github.com/OpenDriveLab/MTGS) | [GTRS](https://github.com/NVlabs/GTRS) | [DiffusionDrive](https://github.com/hustvl/DiffusionDrive) ## ⭐ License and Citation All content in this repository is under the [Apache-2.0 license](https://www.apache.org/licenses/LICENSE-2.0). The released data is based on [nuPlan](https://www.nuscenes.org/nuplan) and is under the [CC-BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/) license. If any parts of our paper and code help your research, please consider citing us and giving a star to our repository. ```bibtex @article{tian2025simscale, title={SimScale: Learning to Drive via Real-World Simulation at Scale}, author={Haochen Tian and Tianyu Li and Haochen Liu and Jiazhi Yang and Yihang Qiu and Guang Li and Junli Wang and Yinfeng Gao and Zhang Zhang and Liang Wang and Hangjun Ye and Tieniu Tan and Long Chen and Hongyang Li}, journal={arXiv preprint arXiv:2511.23369}, year={2025} } ```