| Full text | |
| Author(s): |
Moreira, Daniel A. M.
;
Delgado, Karina Valdivia
;
de Barros, Leliane Nunes
Total Authors: 3
|
| Document type: | Journal article |
| Source: | APPLIED INTELLIGENCE; v. 45, n. 3, p. 662-672, OCT 2016. |
| Web of Science Citations: | 0 |
| Abstract | |
In probabilistic planning problems which are usually modeled as Markov Decision Processes (MDPs), it is often difficult, or impossible, to obtain an accurate estimate of the state transition probabilities. This limitation can be overcome by modeling these problems as Markov Decision Processes with imprecise probabilities (MDP-IPs). Robust LAO{*} and Robust LRTDP are efficient algorithms for solving a special class of MDP-IPs where the probabilities lie in a given interval, known as Bounded-Parameter Stochastic-Shortest Path MDP (BSSP-MDP). However, they do not make clear what assumptions must be made to find a robust solution (the best policy under the worst model). In this paper, we propose a new efficient algorithm for BSSP-MDPs, called Robust ILAO{*} which has a better performance than Robust LAO{*} and Robust LRTDP, considered the-state-of-the art of robust probabilistic planning. We also define the assumptions required to ensure a robust solution and prove that Robust ILAO{*} algorithm converges to optimal values if the initial value of all states is admissible. (AU) | |
| FAPESP's process: | 15/01587-0 - Storage, modeling and analysis of dynamical systems for e-Science applications |
| Grantee: | João Eduardo Ferreira |
| Support Opportunities: | Research Grants - eScience and Data Science Program - Thematic Grants |