Learning-based Vehicle Controller for Vehicle-Pedestrian Interactions

For a detailed description, read our paper located in the report folder. Currently, this work is not published.

Result

Set Up

virtualenv -p python3 myenv
source myenv/bin/activate
pip install gym
cd ./gym-pas
pip install -e .
cd ../
pip install tensorboard
pip install matplotlib
!pip install pyyaml==5.4.1

Install pytorch : https://pytorch.org/get-started/locally/

In 'myenv/lib/python3.6/site-packages/gym/wrappers/order_enforcing.py', add ", collision_flag" in these two lines:

observation, reward, done, info, collision_flag = self.env.step(action)
return observation, reward, done, info, collision_flag

Run the code

source myenv/bin/activate

train python src/main.py

evaluate python src/main_eval.py

Check the training result

tensorboard --logdir ./runs/test --port 6006 or checkout 'checkpoint/output.log'