Improved POMDP Tree Search Planning with Prioritized Action Branching

John Mern, Anil Yildiz, Lawrence Bush, Tapan Mukerji, Mykel J. Kochenderfer

AAAI 2021 pp. 11888-11894

doi:10.1609/AAAI.V35I13.17412 /aaai/2021/mern2021aaai-improved/

Abstract

Online solvers for partially observable Markov decision processes have difficulty scaling to problems with large action spaces. This paper proposes a method called PA-POMCPOW to sample a subset of the action space that provides varying mixtures of exploitation and exploration for inclusion in a search tree. The proposed method first evaluates the action space according to a score function that is a linear combination of expected reward and expected information gain. The actions with the highest score are then added to the search tree during tree expansion. Experiments show that PA-POMCPOW is able to outperform existing state-of-the-art solvers on problems with large discrete action spaces.

PDF AAAI Semantic Scholar

Cite

Text

Mern et al. "Improved POMDP Tree Search Planning with Prioritized Action Branching." AAAI Conference on Artificial Intelligence, 2021. doi:10.1609/AAAI.V35I13.17412

Markdown

[Mern et al. "Improved POMDP Tree Search Planning with Prioritized Action Branching." AAAI Conference on Artificial Intelligence, 2021.](https://mlanthology.org/aaai/2021/mern2021aaai-improved/) doi:10.1609/AAAI.V35I13.17412

BibTeX

@inproceedings{mern2021aaai-improved,
  title     = {{Improved POMDP Tree Search Planning with Prioritized Action Branching}},
  author    = {Mern, John and Yildiz, Anil and Bush, Lawrence and Mukerji, Tapan and Kochenderfer, Mykel J.},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year      = {2021},
  pages     = {11888-11894},
  doi       = {10.1609/AAAI.V35I13.17412},
  url       = {https://mlanthology.org/aaai/2021/mern2021aaai-improved/}
}