Implementation of active inference for highway car following.
- Environment variables are in environment.yml. You might run into an OMP error installing numpy, scipy along with pytorch in anaconda. You can fix this by first installing nomkl (see here).
- Download the INTERACTION dataset and perform the preprocessing steps described in here.
To train the active inference and basedline agents, run:
python ./scripts/train_agent_recurrent.py
You can modify observation features, agent size, learning rate, training epochs by specifying additional arguments. You can use the corresponding .sh script to edit these arguments. Please see the scripts for detailed arguments. To train agents in colab, clone the repo to google drive and run the corresponding .ipynb file.
To test agents on static dataset, run:
python ./scripts/eval_offline.py
To test agents in simulator, run:
python ./scripts/eval_online.py
Description of the simulator can be found here.