This repo contains the 3rd place and the most innovative solution in the 2022 NeurIPS Driving SMARTS Competition. The details of the framework and the prediction network are given here.

conda create -n smarts python=3.8

conda activate smarts

Install SMARTS simulator.

pip install "smarts[camera-obs] @ git+https://github.com/huawei-noah/SMARTS.git@comp-1"

conda install pytorch==1.12.0 -c pytorch

Run train.py in the train folder. Leave the arguments vacant to use the default setting.

python train.py --epochs 1000 --batch_size 64 --learning_rate 2e-4 --device cuda

Run test.py in the submission folder to test the framework with our pretrained model. You can set --envision to visualize the performance of the framework in envision or set --sumo to visualize in sumo.

python test.py --episodes 50 --envisoin

To visualize in Envision (some bugs exist in showing the road map), you need to manually start the envision server and then go to http://localhost:8081/.

scl envision start -p 8081