Experiments Building a Q-table Learner in a Continuous State Space
Translated title
Experimenter omfattende konstruktion af en tabular q-agent i et kontinuerligt miljø
Author
Reffstrup, Jannik Vilhelm
Term
4. term
Education
Publication year
2018
Submitted on
2018-05-31
Pages
14
Abstract
I dette projekt undersøger vi, hvordan man kan opbygge et tabulært Q-læringssystem til problemer med et kontinuerligt tilstandsrum (hvor der er uendeligt mange mulige situationer). Vi identificerer flere praktiske udfordringer og afprøver måder at håndtere dem på. Vores tilgang bruger et sæt 'q-punkter'—punkter, der repræsenterer det kontinuerte tilstandsrum—til at gøre en tabelbaseret lærer mulig. Vi viser, at en sådan lærer kan fungere, men de bedste strategier til at vælge, styre og opdatere disse q-punkter kræver stadig yderligere undersøgelse.
In this project, we explore how to build a tabular Q-learning system for problems with a continuous state space (where there are infinitely many possible situations). We identify several practical challenges and test ways to address them. Our approach uses a set of 'q-points'—points that represent the continuous state space—to make a table-based learner feasible. We show that such a learner can work, but the best strategies for selecting, managing, and updating these q-points still need further investigation.
[This abstract was generated with the help of AI]
Keywords
Documents