This repository provides code exercise and solution for Reinforcement Learning. All code is written in Python and uses custom RL environment from OpenAI Gym.
Goal was train the Reinforcement Learning Drone Agent mastering the flying and accurately hitting a target balloon. A2C was used primaly, PPO and DQN also can be used. Explanations and details are down below ↓
- -score | Proportional to the distance to the target
- Hitting the target is +score, & restart
- Flying out of the game screen is -score & restart
- Agent Y Coordinate
- Angle to Up
- Velocity
- Angular Velocity
- Angle to Target
- Angle Between Target and Velocity
- Distance to Target
Action space is discrete(5). It means there is certain 5 moves the Agent has to do. Do Nothing, Up, Down, Right, Left
- test.py: Testing for the environment. You can display game screen.
- train.py: Trains the Agent. You can change total_steps from Constants.py.
- Agent.py: Drone class.
- evaluate.py: Evaluate the trained model. Elaborated usage is at down below.
- env.py : Environment class. You can alter the game rules, Reward mechanism etc.
- Constants.py : Constant variables of the Game. Screen width, hyperparameters etc.
$ pip install -r requirements.txt
$ python test.py
Default step is 300k. You can alter it from Constants.py
(Loading Libraries may take a while)
$ python train.py
! After the training, your model will be saved in 'models' file. Evaluate your trained model with adding --model parameter to terminal, Or use pretrained models Which in models folder. There is a already pretrained model which named "Crayz_bill" and it is default model.
$ python evaluate.py
If you would like to evaluate your custom model:
$ python evaluate.py --model models/yourmodel
$ python evaluate.py --model models/200k
