Project Goal
Train an agent in RoboCup to learn skills that enable it to score goals past a goalkeeper.
Project steps
-
Read the Deep Robocup paper which doesn't use skills and cannot score past a goalkeeper.
-
Understand the DQN HFO framework
-
Incorporate changes to enable the agent to learn skills
-
Compare the performance with the original Deep RoboCup paper
Required knowledge
-
Random Signals.
-
Any knowledge in machine learning (DL, RL, t-SNE) is an advantage.
Environment
-
C++/python (RoboCup)
-
Torch
Comments and links
Learning to play RoboCup soccer from scratch
using Deep Learning
Department of Electrical Engineering Technion - Israel Institute of Technology
Control Robotics and Machine Learning Lab- המעבדה לבקרה רובוטיקה ולמידה חישובית
Background
RoboCup is a soccer platform that has been used as a testbed for developing Artificial Intelligence algorithms for soccer playing robots. The goal is to have a team of robots that play against the soccer world champions in 2050!
Our goal is to teach an agent to play soccer from scratch. That is, place an agent on a soccer field with a goalkeeper and try and teach him the basic skills which include (1) Approaching the ball, (2) Dribbling to the goalkeeper and (3) scoring past a goalkeeper.