Proceedings

Are White Knights Worth the Trouble? Reconciling POCL and Partial-Order Plans for Plan Optimization

Harrison Oates, Pascal Bercher

Download BibTeX Cite

Proceedings of the Thirty-Fifth International Joint Conference on Artificial Intelligence. Main Track. Pages 6183-6190.

https://doi.org/10.24963/ijcai.2026/688

Keywords

  • Planning and Scheduling: Theoretical foundations of planning

Abstract

Partial order causal link (POCL) planning offers rich structural representations for plan optimization. However, under the classical threat definition, POCL plans form a strict subset of valid partial-order (PO) plans. This gap limits existing POCL optimization techniques to a restricted subspace of PO solutions. We revisit 'white knight' threat semantics, which permit a causal link to persist across a threat provided the deleted condition is re-established by an intermediate producer. We prove that under this definition, the POCL and PO solution spaces become equivalent. By extending a MaxSAT-based optimization framework, we demonstrate that white knights are theoretically necessary for completeness and practically advantageous: they yield plans with significantly fewer ordering constraints in domains with complex causal interference while maintaining computational tractability on standard benchmarks.

Harrison Oates, and Pascal Bercher. 2026. Are White Knights Worth the Trouble? Reconciling POCL and Partial-Order Plans for Plan Optimization. In Proceedings of the Thirty-Fifth International Joint Conference on Artificial Intelligence, IJCAI-26. IJCAI, 6183–6190. https://doi.org/10.24963/ijcai.2026/688
Welcome Future Conferences AI Journal Division IJCAI Proceedings About IJCAI IJCAI Awards