I've been reading about hierarchical reinforcement learning (HRL) and it's applications. A well-written literature review on the subject can be found here. However, I was wondering if research has ever been done where an HRL system has been implemented on an individual robot? This paper seems to imply that it has been, by saying that the delivery task that it models "is commonly used in HRL, both in computational and experimental settings". However, my Google Scholar searches haven't turned up any fruit as to what this real-world experimental setting might be. Help would be appreciated for finding either model-based or model-free implementation of hierarchical reinforcement learning in a robot.
-
$\begingroup$ I'd be very impressed to hear that this was done with a real robot, as the number of trials you need to do for the actual reinforcement is fairly high -- taking significant amounts of time and robot maintenance to complete. $\endgroup$– IanNov 11, 2014 at 22:11
-
$\begingroup$ The University of Southern California had carried out a research. Details can be found in this link perso.ensta-paristech.fr/~stulp/publications/pdfs/… $\endgroup$– DSarkarNov 30, 2014 at 14:53
1 Answer
HRL has been embodied in a robot in multiple cases.
However, how HRL applied in each of these cases varies. The first uses HRL to manipulate Dynamic Movement Primitives, while the second, older method focuses moreso on learning state space values.