3 ms·
The demo GIFs show Sonic 1 and 3, but all 3 Genesis games have slightly different physics and mechanics which could trip up an AI if not trained on all 3 games.
by minimaxir 9y ago
The demo GIFs show Sonic 1 and 3, but all 3 Genesis games have slightly different physics and mechanics which could trip up an AI if not trained on all 3 games. Is the challenge just using the Sonic 1 engine?
- nulltype 9y agoThe challenge is using all 3 Genesis games.
- minimaxir 9y agoYes, just found it in the details: https://contest.openai.com/details https://contest.openai.com/details From the downloaded CSV of game states, all stages from the 3 games are valid training stages, and the custom stages used for testing are derived from all 3 games. Yikes. I really hope those custom stages use the same art assets as the original games.
- oh_sigh 9y agoEvery test run could theoretically start with a jump or some other behavior to exercise the different physics or whatever among the 3 games, and use a discriminator network to feed into one of three networks individually trained on each game?
- tedivm 9y agoThe whole point of the OpenAI exercise is to improve transfer learning- a winning project should be able to work off of a single model. For an example of a system that does learn new rules using a single model check out this post from Vicarious[0]. [0] https://www.vicarious.com/2017/08/07/general-game-playing-with-schema-networks/ https://www.vicarious.com/2017/08/07/general-game-playing-wi...
- taneq 9y agoI'd imagine part of any reinforcement learning approach would be randomly tweaking the physics for each run, to make it robust against small changes and errors. Handling the different games (assuming they're similar enough) should just be an extension of that.