Formally Verified SAT-Based AI Planning

Abstract

We present an executable formally verified SAT encoding of ground classical AI planning problems. We use the theorem prover Isabelle/HOL to perform the verification. We experimentally test the verified encoding and show that it can be used for reasonably sized standard planning benchmarks. We also use it as a reference to test a state-of-the-art SAT-based planner, showing that it sometimes falsely claims that problems have no solutions of certain lengths.

Cite

Text

Abdulaziz and Kurz. "Formally Verified SAT-Based AI Planning." AAAI Conference on Artificial Intelligence, 2023. doi:10.1609/AAAI.V37I12.26714

Markdown

[Abdulaziz and Kurz. "Formally Verified SAT-Based AI Planning." AAAI Conference on Artificial Intelligence, 2023.](https://mlanthology.org/aaai/2023/abdulaziz2023aaai-formally/) doi:10.1609/AAAI.V37I12.26714

BibTeX

@inproceedings{abdulaziz2023aaai-formally,
  title     = {{Formally Verified SAT-Based AI Planning}},
  author    = {Abdulaziz, Mohammad and Kurz, Friedrich},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year      = {2023},
  pages     = {14665-14673},
  doi       = {10.1609/AAAI.V37I12.26714},
  url       = {https://mlanthology.org/aaai/2023/abdulaziz2023aaai-formally/}
}