Revisiting block deordering in finite-domain state variable planning

Article type: Research Article

Authors: Noor, Sabah Binte^{; *} | Siddiqui, Fazlul Hasan

Affiliations: Department of Computer Science and Engineering, Dhaka University of Engineering & Technology, Gazipur-1707, Bangladesh

Correspondence: [*] Corresponding author. E-mail: sabah@duet.ac.bd.

Abstract: Plan deordering removes unnecessary ordering constraints between actions in a plan, facilitating plan execution flexibility and several other tasks, such as plan reuse, modification, and decomposition. Block deordering is a variant of plan deordering that encapsulates coherent actions into blocks to eliminate further ordering constraints from a partial-order plan (POP) and is useful in many applications (e.g., generating macro-actions and improving the overall plan quality). The existing block deordering strategy is formulated in propositional encodings. Finite-domain state variable encodings (e.g., SAS+ representation), in contrast with propositional encodings, can capture the internal structure and the behavior of state variables of a planning instance through concise constructs such as causal graphs (CGs) and domain transition graphs (DTGs). This work redefines the semantics of block deordering terminologies and related plan deordering concepts in finite domain representation (FDR). Our proposed semantics also resolves some limitations of the existing block semantics and further enhance plan flexibility. In addition, this work exploits block deordering to eliminate redundant actions from a POP. A comparative analysis is also performed on block deordering with various deordering/reordering techniques using explanation-based order generalization (EOG) and MaxSAT. Our experiments on the benchmark problems from International Planning Competitions (IPC) show that our FDR formalism of block deordering significantly improves the plan execution flexibility while maintaining good coverage and execution time.

Keywords: Partial-order planning, block deordering, plan deordering, plan annotation, finite-domain representation

DOI: 10.3233/AIC-230058

Journal: AI Communications, vol. 37, no. 4, pp. 563-583, 2024