Relaxed Plan Construction for Metric Optimization in Planning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current plan graph construction methods for metric optimization domains are inefficient when dealing with numeric fluents, as they rely on concurrent properties of time and fail to account for resources like energy use, leading to loose bounds and increased search times.
Innovation Solution
A two-stage relaxed plan construction method that tracks constraints over numeric state variables, allowing for more accurate estimation of metric costs by propagating ranges of consistency and enabling actions, rather than relying solely on maximum values, to avoid irrelevant states and speed up searches.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If current plan graph construction methods use maximum time propagation for numeric fluents, then the method is simple to implement, but the bound on metric cost becomes loose and search time increases
Solution Approach 1:
The patent changes the parameter for propagating numeric fluents from maximum time to sum of time. Specifically, when computing the earliest time a fact can be achieved, the system sums the time durations of actions required to establish preconditions rather than taking the maximum, thereby tightening the metric bound while maintaining admissibility
Solution Approach 2:
The patent applies a relaxed plan construction that intentionally overestimates the metric cost by using sum propagation instead of maximum propagation. This excessive action creates a tighter bound that guides the search more effectively, reducing the need to explore irrelevant states while maintaining computational tractability
2Productivity
If sum propagation is used to tighten the bound on metric cost, then search efficiency improves, but the method becomes more complex and may overcount resource usage
Solution Approach 1:
The patent implements feedback mechanisms where the relaxed plan's metric cost is used to guide the search process. The system continuously refines the bound based on the actual cost of actions taken, allowing the search to focus on promising paths while maintaining admissibility through the relaxed planning graph
Solution Approach 2:
The patent segments the planning problem into two parts: a relaxed plan construction phase that computes tight bounds using sum propagation, and a search phase that uses these bounds to guide action selection. This segmentation allows the complex sum propagation to be performed once during relaxed planning without burdening the entire search process
3Productivity
If relaxed plan construction ignores delete effects, then computation becomes tractable, but accuracy of metric estimation decreases
Solution Approach 1:
The patent extracts the delete effects from the planning graph construction, focusing only on add effects for computing the relaxed plan bound. By separating the computation of metric bounds (which uses only add effects) from the full planning problem (which must handle delete effects), the system achieves tractable computation while maintaining sufficient precision for guiding the search
Data Source
AI summary
Methods and apparatus to provide techniques for planners that records constraints over a set of numeric state variables for each fact and characterizes the set of states in which that fact is true to provide more information to reason about mutual exclusions among sets of states offering a tighter bound on the metric cost of a plan than existing planners currently attain. This more accurate estimate can avoid irrelevant states and decrease search times.