The Metric-FF Planning System: Translating "Ignoring Delete Lists" to Numeric State Variables (1106.5271v1)

Abstract: Planning with numeric state variables has been a challenge for many years, and was a part of the 3rd International Planning Competition (IPC-3). Currently one of the most popular and successful algorithmic techniques in STRIPS planning is to guide search by a heuristic function, where the heuristic is based on relaxing the planning task by ignoring the delete lists of the available actions. We present a natural extension of ignoring delete lists'' to numeric state variables, preserving the relevant theoretical properties of the STRIPS relaxation under the condition that the numeric task at hand ismonotonic''. We then identify a subset of the numeric IPC-3 competition language, ``linear tasks'', where monotonicity can be achieved by pre-processing. Based on that, we extend the algorithms used in the heuristic planning system FF to linear tasks. The resulting system Metric-FF is, according to the IPC-3 results which we discuss, one of the two currently most efficient numeric planners.

• The paper introduces a heuristic extension that adapts the FF planning framework to handle numeric state variables by ignoring negative effects.
• The paper preprocesses numeric tasks to identify linear and monotonic scenarios, enabling efficient application of the extended algorithm.
• The paper demonstrates Metric-FF’s competitive performance in planning competitions by achieving polynomial-time tractability in numeric domains.

An Examination of the Metric-FF Planning System for Numeric State Variable Planning

The paper under discussion explores the extension of heuristic functions from the STRIPS planning paradigm to numeric state variables. The authors propose the Metric-FF planning system, which builds upon the FF heuristic framework by adapting it to handle numeric constructs. This adaptation is guided by the need to navigate planning tasks involving numeric state variables, particularly focusing on conditions where tasks exhibit monotonic properties.

Background and Motivation

The research is grounded in the context of the third International Planning Competition (IPC), highlighting the challenges faced when extending planning systems beyond STRIPS to accommodate numeric constructs. Traditional approaches often relied on ignoring the delete lists to simplify planning tasks. This paper introduces a parallel technique applicable to numeric state variables while preserving theoretical properties, specifically in monotonic scenarios.

Key Contributions

1. Heuristic Extension to Numeric Variables: The paper presents a heuristic framework that ignores negative numeric effects. This extension maintains admissibility, informedness, and tractability within polynomial time under certain conditions.
2. Monotonicity and Linear Task Identification: By preprocessing numeric tasks, the authors identify linear tasks that can be expressed in a monotonic fashion. This approach allows them to apply the extended FF algorithms to these preprocessed tasks, demonstrating practical efficiency.
3. Implementation of Metric-FF System: The paper introduces Metric-FF, extending the FF planning system to handle linear tasks with numeric state variables. It operates effectively by using a Graphplan-inspired strategy to construct relaxed plans, adapting previous FF techniques to numeric contexts.
4. Performance Evaluation: The system's performance is evaluated in numeric domains from IPC, showing competitive efficiency, particularly in tasks with numeric state variables that are inherently linear and monotonic after preprocessing.

Theoretical and Practical Implications

From a theoretical perspective, the research identifies the properties that ensure the heuristic's effectiveness, focusing on monotonicity and linear expressibility. This provides a framework for extending other heuristic-based systems to numeric planning problems. Practically, the implementation of Metric-FF demonstrates a feasible method for dealing with numeric constructs by leveraging preprocess transformations and established heuristic methods applied to STRIPS.

Future Directions

The paper suggests avenues for future research in extending the existing framework to more complex numeric languages that transcend linear expressibility. Investigating languages with richer expressions and identifying weaker criteria for monotonicity and tractability could further enhance the generalizability and efficiency of planning systems like Metric-FF.

In conclusion, this paper provides significant insights into extending heuristic planning to domains with numeric state variables. While the Metric-FF system showcases high efficiency in specific cases, ongoing exploration into broader language support and optimization considerations remains crucial to fully unlocking its potential across diverse real-world applications.