This repo is the implementation of the following paper:
** Ke Guo, Wei Jing, Junbo Chen, Jia Pan. CCIL: Context-conditioned imitation learning for urban driving. RSS, 2023**
Download Lyft's Python software kit.
Download the Lyft Motion Prediction Dataset; only the files in Training Dataset(8.4GB), validation Dataset (8.2GB), Aerial Map and Semantic Map
are needed.
Store all files in a single folder to match this structure: https://woven-planet.github.io/l5kit/dataset.html.
Download nuPlan's Python software kit.
Download the nuPlan-v1.0 Dataset; only the files in Maps, Val Split, Test Split and Train Split for Las Vegas City
are needed.
Store all files in a single folder to match this structure: https://github.com/motional/nuplan-devkit/blob/master/docs/dataset_setup.md.
Run train.py
to learn the planner. You need to specify the model name --model_name
and the file paths to dataset --data_root
. Leave other arguments vacant to use the default setting.
python train.py \
--name lyft \
--data_root /path/to/lyft/data \
Run eval.py
to do closed-loop testing. You need tospecify the model name --model_name
and the file paths to dataset --data_root
. Leave other arguments vacant to use the default setting.
python eval.py \
--name lyft \
--data_root /path/to/lyft/data \
More pretrained model, visualization, data information can be found in Google Drive.