To get required python packages on MacOS
pip install -r requirements.txt
If you use other platform or the above command fails, basically you need to install pygame,numpy for this simulation. And if you want ot use deep q learning in deep_q_learning.py, please install pytorch, torchvision, opencv to train the network, please refer to here to install pytorch and torchvision.
To run the simulation
python q_learning.py test
To train the model
python q_learning.py train PATH_TO_SAVE_MODEL
To load a pretrained_model, and let the player play
python q_learning.py play PATH_TO_MODEL
The simulator is in game/, to build a simulator:
import sys
sys.path.append("PATH_TO_GAME_FOLDER")
import deep_traffic as game
s = game.GameState()
To interact with the simulator:
do_nothing = 0
image, reward, terminate, (x,y), up, red , speed, pedes = s.frame_step(do_nothing)
Where image is the current screen of simulator, reward is current reward, terminate indicate whether game is over, (x, y) is the location of player, red means whether the corresponding traffic light is red, speed is the current speed of agent, and pedes is the location of pedestrian. The input argument is the action, which is 0 in above example, indicating do nothing. The other actions are:
1 : left shfit lane
2 : right shift lane
3 : speed up
4 : slow down
To train the model, need torch and pytorch installed, you may look reference here
The AI of the game is White car, they will avoid crash and speed up in a smart way. They may also park or restart from parking. When parking, the AI car becomes yellow. In the environment, there is also traffic light and pedestrian. Everyone should obey traffic light, and pedestrian can across the road at any time, any where. Every car should not hurt pedestrian.
To Do
- Design reward system to make better player
- Make simulator stable
- Reduce library dependency