News for hackers. Stuff that matters.
Re:Algorithms for Reinforcement Learning by clickok on Monday February 18, 2019 @10:31PM (#19196393)
Yes.

I don't know what your thresholds for impressive might be, or what your basis of comparison is, but pure control theory seems to be at a bit of a dead end, whereas reinforcement learning allows for greater flexibility and robustness for control tasks, subject to your willingness to gather (or simulate) a lot of training data[0].

See for example this video[1] in which OpenAI shows off a robotic arm with "unprecedented" dexterity.

-----

0. https://www.youtube.com/watch?v=ZVIxt2rt1_4 for a video (there's also an associated paper) about setting up RL tasks for robots.

1. https://www.youtube.com/watch?v=jwSbzNHGflM

There are no comments on this comment yet.