Atari-freeway is a game that a player can control a chicken to get to the other side of a ten-lane highway with heavy traffic flow. Once hit by a car, the chicken will be forced back slightly. While dodging the vehicles, we have to make it reach to the endpoint as fast as it can since our goal is to make the chickens that cross the road safely as many as possible within a limited time. That is, we must to train our chicken to march in an efficient way.
Among the algorithms in reinforcement learning, we chose Q-learning and DQN to train our agent. Thus we can improve it by using DQN to approximate Q-function by neural network.
Run the following command to set the environment of Atari Freeway
pip install gym[atari]
and
pip install ale-py
Run the following command to perform Q-learning training process:
python Qlearning_train.py
Run the following command to perform DQN training process:
python DQN_train.py
Run the following command to modify and check the imformation of parameters:
python DQN_train.py -h and python Qlearning_train.py -h
| 陳尚奇 109550110 | 曾偉杰 109550156 | 李驊恩 109550159 | |
|---|---|---|---|
| Brainstorming (10%) | 33.3% | 33.3% | 33.3% |
| Latex Report (15%) | 0% | 0% | 100% |
| DQN Training Code (20%) | 20% | 80% | 0% |
| Q-learning Training Code (15%) | 70% | 30% | 0% |
| Model Training and Data Collection (15%) | 40% | 30% | 30% |
| Video Recording (10%) | 100% | 0% | 0% |
| Data and Result Analysis (15%) | 10% | 80% | 10% |