GRAPH-CONSTRAINED DIFFUSION FOR END-TO-END PATH PLANNING

Data preparison

Make directories as below

sets_data/
  |--real/
    |--map/
    |--raw/
    |--trajectories/
  |--real2/
    |--map/
    |--raw/
    |--trajectories/

Then download original data from https://gaia.didichuxing.com/ and the GPS file under directory raw/. I have put a small sample file in raw/ for better understanding, simply delete the suffix "_small" to use it. Then execute

python -m loader.preprocess.mm.process_all

Model Training

Execute

bash ./train.sh

The model will be found at sets_models/ The evaluation results will be found at sets_res/