Current students


DRAPPO GIANLUCACycle: XXXVII

Section: Computer Science and Engineering
Advisor: RESTELLI MARCELLO
Tutor: GATTI NICOLA

Major Research topic:
Hierarchical Reinforcement Learning

Abstract:
Hierarchical Reinforcement Learning (HRL) is the sub-framework of Reinforcement Learning that enables solving complex and temporally extended tasks. When dealing with real-world applications, most of the time an artificial agent needs to plan considering a huge amount of action far in the future, which enlarges the problem complexity for standard Reinforcement Learning approaches that normally suffer in long-horizon planning. HRL, on the other hand, can decompose the problem into a bunch of subproblems organized in a hierarchy, reducing the single problem horizon and hence complexity. Afterwards, once all the subproblems are solved, the solution can be stitched together to provide a solution to the original bigger task.This thesis will analyse how and when can be beneficial to use HRL instead of standard RL techniques, considering different scenarios such as multi-agent ones, or others in which the reward function needs to be inferred.