Usage

Installation

1. Install SUMO (ideally 1.12, but 1.13 seems to work as well) according to your OS, and make sure that the env var SUMO_HOME is set.

2. Create a conda env or venv with python 3.7

3. Install dependencies using following command:

pip install -r requirements.txt

Run Instructions

<dir> is where all the training artifacts will be stored and/or the checkpoints will be retrieved (to evaluate or restart the training). Will be created if it doesn’t exist.

From the root of the repo and the correct venv/ conda env, run the following command to train the agents. Check code/train.py for more options.

python code/train.py --dir <exp_dir>

python code/train.py --dir <exp_dir> --kwargs <python dict with arguments to override the config in main.py>

Example:

python code/train.py --dir wd/test --kwargs "{'wandb_proj':'scenario-env'}"

From the root of the repo and the correct venv/ conda env, run the following command to evaluate the trained agents on a given intersection dataset. Check code/evaluate.py for more options. Note that <dir> should point to the same directory used for training.

python code/evaluate.py --dir <exp_dir>

From the root of the repo and the correct venv/ conda env, run the following command to visulize the trained agents on a given intersection dataset. Check code/visualize.py for more options. Note that <dir> should point to the same directory used for training.

python code/visualize.py --dir <exp_dir>