This project will investigate automatic model generation of one of the most general frameworks for sequential decision-making under uncertainty, namely the Partially Observable Markov Decision Processes (POMDPs). Specifically, we will investigate how LLMs can help generate POMDP models, represented as codes, such as [1]. The work itself will involve exploring the potential integration of [1] and our recently developed POMDP solver [2].
References:
[1] Curtis, A., Tang, H., Veloso, T., Ellis, K., Tenenbaum, J. B., Lozano-Pérez, T., & Kaelbling, L. P. (2025, October). LLM-guided probabilistic program induction for POMDP model estimation. In Conference on Robot Learning (pp. 3137-3184). PMLR. [https://proceedings.mlr.press/v305/curtis25a.html]
[2] Hoerger, M., M. Sudrajat, and H. Kurniawati. “Vectorized Online POMDP Planning.” arXiv preprint arXiv:2510.27191 (2025). [https://arxiv.org/abs/2510.27191]
Requirements:
- Have a good understanding of basic probability [Must]
- Fluent in programming with Python [Must]
- Received a mark of >= 75 in Algorithms (COMP3600/6466) OR Intelligent Robotics course (COMP4620/8620) in 2023 / 2025 OR COMP4680/8650 in 2024 [Good to have]
- Have a GPA >= 6.5/7.0 [Good to have]
If you are interested, please send me an email to schedule a chat, and please send me a copy of your CV and transcript.