Autor: |
Gottesman, Omer, Asadi, Kavosh, Allen, Cameron, Lobel, Sam, Konidaris, George, Littman, Michael |
Rok vydání: |
2021 |
Předmět: |
|
Druh dokumentu: |
Working Paper |
Popis: |
Principled decision-making in continuous state--action spaces is impossible without some assumptions. A common approach is to assume Lipschitz continuity of the Q-function. We show that, unfortunately, this property fails to hold in many typical domains. We propose a new coarse-grained smoothness definition that generalizes the notion of Lipschitz continuity, is more widely applicable, and allows us to compute significantly tighter bounds on Q-functions, leading to improved learning. We provide a theoretical analysis of our new smoothness definition, and discuss its implications and impact on control and exploration in continuous domains. |
Databáze: |
arXiv |
Externí odkaz: |
|