search

pocokhc / simple_distributed_rl

Simple Distributed Reinforcement Learning Framework（シンプルな分散強化学習フレームワーク）
MIT License
41 stars 6 forks source link

General Question on Agent57 #14

Closed hlsafin closed 1 year ago

hlsafin commented 1 year ago

I know that you mentioned on your blog that you are limited by your compute power; I guess I am just asking if you replicated any results from the Agent57 paper, especially when it comes to hard exploration games like Montezuma's revenge, did you see any early signs of good results on this game? Thank you

pocokhc commented 1 year ago

thanks for your questions. The results of the paper have not been reproduced. We have confirmed that it is possible to learn simple games, but we have not confirmed that games with sparse rewards can be learned.