Author: iadine.chades@csiro.au 8/07/2020
Partially Observable Markov Decision Processes (POMDPs) are a convenient mathematical model to solve sequential decision-making problems under imperfect observations. The POMDPproblems repository aim to provide examples of POMDP problems published or unpublished in ecology, conservation, biosecurity, and health (list to be completed). We hope to inspire further collaborations between domain experts and the AIML community. The challenges we are interested in solving are:
- Improve POMDP expertise across domain experts;
- Increase interpretability and explainability of POMDP solutions;
- Improve scalability and algorithms performances.
We are currently limiting our repository to the following POMDP file formats: Tony's pomdp format (.pomdp), SPUDD-SymbolicPerseus factored representation (.txt) and XML SARSOP (.pomdpx). If you wish to contribute to this repository please follow these steps:
- Insure a POMDP problem can be shared (license, approval, acknowledgments etc);
- Please describe the problem and reference to published literature;
- Submit contribution;
The submission will be tested on one of the following solvers:
- pomdp-solve http://www.pomdp.org/code/
- perseus http://www.st.ewi.tudelft.nl/mtjspaan/software/
- symbolic perseus https://cs.uwaterloo.ca/~ppoupart/software.html
- SARSOP https://github.com/AdaCompNUS/sarsop
If you are using this repository please cite: Chadès, I., Pascal, L. V., Nicol, S., Fletcher, C. S., & Ferrer-Mestres, J. (2021). A primer on partially observable Markov decision processes (POMDPs). Methods in Ecology and Evolution, 12, 2058– 2072. https://doi.org/10.1111/2041-210X.13692
or see cite this repository on the right panel.