OpenAI has launched a transfer learning contest using the Sonic The Hedgehog series of games for SEGA Genesis. The challenge is to create the best agent for playing custom levels of the Sonic games — without having access to those levels during development.

The contest uses Gym Retro, a wrapper for video games emulator cores that  includes support for multiple classic game consoles and a dataset of different games including 30 SEGA Genesis games.

Gym Retro is OpenAI's second generation attempt to build a large dataset of reinforcement learning environments. It builds on Universe, which we reported when it was released in 2016 which, because its environments ran asynchronously, could only run in real time, and suffered unreliability due to screen-based detection of game state. Gym Retro extends the model of the Arcade Learning Environment to a much larger set of potential games.

Introducing the contest the OpenAI blog explains:

In typical RL research, algorithms are tested in the same environment where they were trained, which favors algorithms which are good at memorization and have many hyperparameters. Instead, our contest tests an algorithm on previously unseen video game levels. 

Credit: OpenAI blog 

The contest, which opened on April 5th and runs until June 5th, provides a training set of levels from Sonic The Hedgehog games and evaluates contestants' algorithms on a test set of custom levels created for the contest. To help you get started, OpenAI has released several RL algorithms that you can tweak in a retro-baselines repo on GitHub. 

As you can see from the baseline results, RL algorithms fall well below human performance - the red dashed line - even though the human's played for one hour versus eighteen for the algorithms.

Some of the baseline results, and the new benchmark intended to measure the performance of transfer learning and few-shot learning algorithms in the RL domain, are detailed in a technical paper  from Open AI.

To take part in the contest you'll need to register with OpenAI and also have a Steam username and password. Details of setting up Gym Retro and creating a simple agent in Python are given on the Retro Contest page. The leaderbord currently has two clusters of scores - five teams with scores around 3,700 and another 15 around 13,500. But with eight weeks to go, it is all still to play for - or should that be to train for.

There are two award categories, "Best Score" and "Best Writeup". To be eligible to win you must release your submission as open source at the end of the contest. 1st, 2nd, and 3rd place winners from each category will receive a trophy. In addition there will be a single award for "Best Supporting Materials". All winners will be invited to co-author a tech report with OpenAI about the contest.

More Information

OpenAI Retro Contest

Gotta Learn Fast: A New Benchmark for Generalization in RL

Related Articles

OpenAI Gym Gives Reinforcement Learning A Work Out

AI Goes Open Source To The Tune Of $1 Billion

OpenAI Universe - New Way of Training AIs

To be informed about new articles on I Programmer, sign up for our weekly newsletter, subscribe to the RSS feed and follow us on Twitter, Facebook or Linkedin.

Comments

Make a Comment or View Existing Comments Using Disqus

or email your comment to: comments@i-programmer.info