Reinforcement Learning - 2019 Assignment

Environment

Unity Obstacle Tower [repo link]

Packages

• torch
• torchvision
• gym
• matplotlib
• mlagents (v0.10.1)
• pillow
• opencv-python

Create Conda Env from environment.yml

conda env create -f environment.yml -n assignment

Evaluation - MyAgent

• The MyAgent will be evaluated
• Initialize model within the __init__ method
• Return action from act(observation) method

Run Docker for Evaluation

• When initializing ObstacleTowerEvaluation env, set docker_training=True

    cd to_root_folder
  
    docker build -t assignment .
  
    docker run -it --rm assignment
  

Record Agent

• Install ffmpeg [Ubuntu] [macOS] [Windows]

    python recorder.py