Optimal Cost-Bounded Plan Reduction via ILP
This study tackles the challenge of deriving a valid subplan from a pre-existing plan that maximizes utility within a specified cost limit, adhering to the requirement of maintaining the sequence and integrity of original actions. Each goal possesses a designated utility value; thus, the plan is streamlined by eliminating actions linked to lower-utility goals while ensuring that executability and the original order of actions are intact. The decision variant is proven to be NP-complete. Two precise approaches are introduced: one through oversubscription planning (OSP) and the other via Integer Linear Programming (ILP). This research builds on earlier findings presented by Del Toro, Fuentetaja, and García-Olaya at ICAPS 2026, offering an enhanced ILP formulation that significantly minimizes model size.
Key facts
- Problem: extract valid subplan from precomputed plan maximizing utility under cost bound
- Constraints: only original actions and their order allowed
- Each goal assigned a utility value
- Plan reduced by removing actions supporting low-utility goals
- Decision variant is NP-complete
- Two exact methods: OSP and ILP
- Extends previous work at ICAPS 2026
- Refined ILP formulation reduces model size
Entities
—