Skip to content

Our NIPS 2017 “Learning to Run” Approach

medium.com
42 pointsmarcelsalathe4 comments
On HN

Comments

That simulation reminds me of the flash game 'QWOP'

Had the same thought. :) I wonder if anyone has done a challenge using that game which has simpler inputs I believe?

Simplest RL algorithm (Q-learning) achieves 100m in QWOP: https://www.youtube.com/watch?v=e27TUmMkOA0

Although it found and exploited a local maximum of "knee scraping" technique (which humans can replicate) :)

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.