Department of Electrical Engineering
Control Robotics and Machine Learning Lab
Technion - Israel Institute of Technology
המעבדה לבקרה רובוטיקה ולמידה חישובית
Playing Football with Deep Reinforcement Learning
Background
Deep Reinforcement Learning has seen multiple successes over the recent years. It begun with the ATARI game simulator, when Google presented the Deep Q learning paper, and has recently evolved to much more complex and dynamic domains.
In this work, we will focus on the Google Research Football domain, a new domain recently introduced by Google. The game provides several forms of information, (1) a pixel based view of the game, (2) a mini-map view of the game and (3) a vector form representing the location of objects, in addition to static information.
Project Goal
-
Learn how to work with the Google Research Football domain.
-
Invent methods for learning policies with tunable performance.
-
Evaluate against real human players.
-
Publish results.
Project steps
-
Understand the DQN framework.
-
Understand the Google Research Football domain.
-
Suggest your own ideas!
Required knowledge
-
Strong programming skills.
-
Knowledge in Deep Learning.
-
Any knowledge in RL is an advantage.
Environment
-
Python
-
PyTorch.
Comments and links
-
See the DQN paper by Google.
-
See the Google Research Football paper.