Data Preprocess

Step 1. Download data

• Download the nuPlan dataset from the official website and structure it in the recommended format. Link the dataset under data.
• Download our custom AV behaviors from Hugging Face to data/custom_behaviors.
• Download checkpoints from Hugging Face to ckpts.

Step 2. Set up the data preprocessing environment

cd reactsim_data_pipeline
conda create -n reactsim_dp python=3.9 pip -y
conda activate reactsim_dp
pip install -r requirements.txt

Step 3. Run data preprocessing

bash preprocess.sh

After preprocessing, the directory structure will look like:

ReactSim-Bench/
├── ckpts/
|   ├── clsft.ckpt
|   ├── ctg.ckpt
|   └── ...
├── CTG/
├── data/
|   ├── custom_behaviors
|   |   ├── sg-one-north_0b7ca3ca037552bf_adb475ee1f565362.pkl
|   |   └── ...
|   ├── motion_anchors
|   |   ├── agent_vocab_555_s2_nuplan.pkl
|   |   └── ...
|   ├── nuplan
|   |   ├── maps/
|   |   └── nuplan-v1.1/splits/trainval
│   |       ├── 2021.05.12.22.00.38_veh-35_01008_01518.db
|   |       └── ...
|   └── nuplan_processed
|       ├── train/processed_scenarios/
│       │   ├── sg-one-north_0a0c5449b9905073_5fcefba67eb95e5b.pkl
|       │   └── ...
|       └── val/processed_scenarios/
│           ├── sg-one-north_0b7ca3ca037552bf_adb475ee1f565362.pkl
|           └── ...
├── docs/
├── MTR/
├── reactsim_data_pipeline/
├── SMART_CATK_TrajTok/
└── VBD/