Generative Adversarial Imitation Learning on NGSIM I-80 Dataset

This repository contains three components: an improved version of ILSwiss, which is an RL training suite containing the implementation of Generative Adversarial Imitation Learning (GAIL), and two environments. The two environments both are driving simulators that use NGSIM I-80 dataset as background traffic. One of them is PPUU, the other is implemented as an imitation learning scenario based on SMARTS platform.

How to setup:

0. Clone this repository

1. Install python requirements with

pip install -r requirements.txt

Setup SMARTS

2. Download the SMARTS platform as a submodule with

   git submodule init
   git submodule update

3. Install SMARTS simulation platform in the ./SMARTS folder following its instructions

   cd ./SMARTS
   bash ./utils/setup/install_deps.sh
   echo "export SUMO_HOME=/usr/share/sumo" >> ~/.bashrc
   pip install -e .[camera-obs]

4. Download the NGSIM dataset from: https://drive.google.com/file/d/1SILBfK2Z1LiTJ9cc4qGErq547Ax5nX8M/view?usp=share_link, and place it under smarts-imitation/

5. Install the ngsim wrapper for smarts with

   pip install -e ./smarts-imitation
   # You may encounter the ModuleNotFoundError when running this command,
   # which is okay. Please just wait until it terminates.
   scl scenario build --clean ./smarts-imitation/ngsim

6. Generate structured expert demonstrations with

   cd ILSwiss
   python run_experiment.py -e exp_specs/gen_expert/smarts.yaml

7. Run GAIL with

   python run_experiment.py -e exp_specs/gail/gail_smarts.yaml

Setup PPUU

8. Make sure you have downloaded the NGSIM dataset and placed it under smarts-imitation. Create a link to the dataset under PPUU folder:

   mkdir pytorch-PPUU/traffic-data
   ln -s $(pwd)/smarts-imitation/xy-trajectories pytorch-PPUU/traffic-data/

   Process the traffic data to dump the "state, action, cost" triple:

   cd pytorch-PPUU
   for t in 0 1 2; do python generate_trajectories.py -map i80 -time_slot $t; done

9. Generate structured expert demonstrations with

   python run_experiment.py -e exp_specs/gen_expert/ppuu.yaml

10. Run GAIL with

python run_experiment.py -e exp_specs/gail/gail_pppu.yaml