top of page
Department of Electrical Engineering
Control Robotics and Machine Learning Lab
Technion - Israel Institute of Technology
המעבדה לבקרה רובוטיקה ולמידה חישובית

The OpenAI Universe Challenge

Background
 

OpenAI recently released a set of over 1000 domains called Open AI Universe. Example domains include 3D racing games as well as real-world tasks such as booking a flight online. The idea is to train agents to solve these domains using Reinforcement Learning (RL) techniques; That is, train an agent to race a car and win the game or train an agent to book a flight online. 

​

Deep Reinforcement Learning algorithms have enabled RL to scale to solving much larger domains. These include playing Atari video games (better than humans in many cases), beating the world champion at Go, grasping and manipulation in robotics etc.

Project Goal

 

Many of the domains in OpenAI have not as yet been benchmarked. That means that they have not yet been solved. The goal of this project is to choose a challenging domain (e.g., a 3D racing game) and try to solve it using the state-of-the-art Deep RL techniques. 

Project steps

 

  • Learn about the various Deep RL algorithms and architectures

  • Choose a domain from OpenAI Universe

  • Learn about the OpenAI framework

  • Solve the domain using Deep RL

​

Required knowledge

​

  • Strong programming skills - Especially python

  • Any knowledge in DL and RL is an advantage. 

​

Environment

  • TensorFlow.

​

Comments and links

​

 

bottom of page