/Dragtraffic

Repo for DragTraffic: Interactive and Controllable Traffic Scene Generation for Autonomous Driving.

Primary LanguagePython

Dragtraffic

The Repo for Dragtraffic: Interactive and Controllable Traffic Scene Generation for Autonomous Driving.

News: Dragtraffic has been accepted by IROS2024, The training and generation code is coming soon.

Webpage | Paper |

We are preparing

We are sorting out the code and the relevant code will be released soon in the following order:

    • Paper releasing(Accepted).
    • Model files and checkpoints.
    • UI and Inference demo code.
    • Training code.

Setup environment

# Clone the repository and set up the environment:

git clone https://github.com/chantsss/Dragtraffic
cd dragtraffic

# Create a virtual environment
conda create -n dragtraffic python=3.9.12
conda activate dragtraffic

# Install PyTorch and other dependencies
pip install torch==1.13.0+cu116 torchvision==0.14.0+cu116 torchaudio==0.13.0 --extra-index-url https://download.pytorch.org/whl/cu116 
pip install -e .

Test the setup with the following scripts. These do not require downloading additional data.

# Vehicle Placement Model
# Replace the data path in dragtraffic/init/configs/local.yaml
python train_init.py -c local
# Trajectory Generator Model
# The same to dragtraffic/act/configs/local.yaml
python train_act.py -c local

Quick Start

Download dataset for road and traffic

Download from Waymo Dataset

Note: it is not necessary to download all the files from Waymo. You can download one of them for a simple test.

Data Preprocess

# python -m pdb label_trans20.py path_A path_B None
mkdir PATH_B
python -m pdb label_trans20.py /media/sheng/ssd/dataset/waymo/training_20s path_B None --debug_vis --draw_scene_pi 

Specify 'debug_vis' 'draw_scene_pi' to see the data visualization

Train Dragtraffic

# Train
python -m pdb train_act.py -c local -d 0 -e quick_test -ca 1 -ei 10

Specify  '-d' for devices '-ca' for using cache, '-ei' means evaluation interval 

# Note that we found pretrain init model and diff model then finetune them together in led model.

1. Train the init model by setting model:'drag' in act/local.yaml
2. Train the diff model by setting model:'diffusion' in act/local.yaml
3. Train dragtraffic model by setting setting model:'led', init_model_path and diff_model_path obtained from step 1&2.


# Eval
python -m pdb train_act.py -c local  -d 0 -e drag_test_eval -ca 1 -ev 1 -mp MODEL_PATH

Specify  '-ev' 1 '-mp' means model path

Generate new traffic scenarios

# Running following scripts will generate images and GIFs 

# Scene editing, insertion based on UI
# Refer to gradio_visualizer.py to see the hyperparameter
python gradio_visualizer.py
# Note if got unexpected keyword argument 'socket_options', try pip install --upgrade httpx httpcore

# Scene editing, insertion based on the dataset
python -m pdb ui_utils_moe.py -l 90 -p 30 -gn 10
# Use act model to generate gif, -l means length, -p means pertime length, -gn means generate nums 
# Note here you should specify the act model address in init config

Acknowledgements

This repository is built on the following open source projects, many thanks to them for their great work:

Citation

If Dragtraffic is helpful in your research, please consider giving us a star or citing our work:

@misc{wang2024dragtraffic,
      title={Dragtraffic: A Non-Expert Interactive and Point-Based Controllable Traffic Scene Generation Framework}, 
      author={Sheng Wang and Ge Sun and Fulong Ma and Tianshuai Hu and Yongkang Song and Lei Zhu and Ming Liu},
      year={2024},
      eprint={2404.12624},
      archivePrefix={arXiv},
      primaryClass={cs.RO}
}