Abstract:Graph learning is naturally well suited for use in symbolic, object-centric planning due to its ability to exploit relational structures exhibited in planning domains and to take as input planning instances with arbitrary numbers of objects. Numeric planning is an extension of symbolic planning in which states may now also exhibit numeric variables. In this work, we propose data-efficient and interpretable machine learning models for learning to solve numeric planning tasks. This involves constructing a new graph kernel for graphs with both continuous and categorical attributes, as well as new optimisation methods for learning heuristic functions for numeric planning. Experiments show that our graph kernels are vastly more efficient and generalise better than graph neural networks for numeric planning, and also yield competitive coverage performance compared to domain-independent numeric planners. Code is available at <a class="link-external link-https" href="https://github.com/DillonZChen/goose" rel="external noopener nofollow">this https URL</a>

What problem does this paper attempt to address?

The core problem that this paper attempts to solve is: **How to use graph - learning methods to learn solution strategies efficiently and interpretably in Numeric Planning tasks**. Specifically, the author focuses on how to handle graphs with both continuous and categorical attributes by constructing new graph kernels and optimization methods, and apply them to heuristic function learning in numeric planning. ### Main Research Questions 1. **Challenges in Numeric Planning Tasks**: - Numeric planning is an extension of classical planning. States can contain numerical variables, which makes the state space more complex. - The undecidability of numeric planning tasks (undecidable in the general case) makes them more difficult to solve than classical PSPACE - complete planning. 2. **Limitations of Existing Methods**: - Traditional symbolic AI methods are difficult to handle large - scale numeric planning tasks. - Although deep - learning methods perform well in some fields, they may perform poorly in numeric planning tasks due to problems such as over - fitting. 3. **Proposed New Methods**: - Construct a new graph kernel (CCWL kernel) that can handle both continuous and categorical attributes in graphs simultaneously. - Propose a framework (GOOSE) based on classical machine learning and deep learning for learning heuristic or value functions in numeric planning tasks. - Introduce a new ranking formula for learning heuristic or value functions to improve the generalization ability and efficiency of the model. ### Main Contributions of the Paper - **New Graph Representation**: Introduce the Numeric Instance Learning Graph (νILG) for representing numeric planning tasks, which is suitable for classical and deep graph - learning methods. - **Extended WL Kernel**: Propose the CCWL kernel, which can meaningfully handle graphs with continuous and categorical attributes. - **New Ranking Formula**: Introduce a new linear - programming method for learning heuristic or value functions, which solves some problems in existing methods. ### Experimental Results Experiments show that the graph - kernel method proposed by the author is more efficient than Graph Neural Networks (GNNs) in numeric planning tasks, and is comparable to domain - independent numeric planners in coverage performance. ### Summary This paper aims to improve the solution efficiency and interpretability of numeric planning tasks through graph - learning methods, especially by constructing new graph kernels and optimization methods to deal with complex tasks with both continuous and categorical attributes.

Graph Learning for Numeric Planning

Graph Learning for Planning: The Story Thus Far and Open Challenges

Can Graph Learning Improve Planning in LLM-based Agents?

Learning Domain-Independent Heuristics for Grounded and Lifted Planning

Learning General Optimal Policies with Graph Neural Networks: Expressive Power, Transparency, and Limits

Learning to plan with uncertain topological maps

-Equivariant Graph Planning for Navigation

Graph-enhanced Large Language Models in Asynchronous Plan Reasoning

Neural-Guided RuntimePrediction of Planners for Improved Motion and Task Planning with Graph Neural Networks

Graph schemas as abstractions for transfer learning, inference, and planning

Return to Tradition: Learning Reliable Heuristics with Classical Machine Learning

GraphPlanner: Floorplanning with Graph Neural Network.

E(2)-Equivariant Graph Planning for Navigation

Cooperative Behavior Planning for Automated Driving Using Graph Neural Networks

Sparse Graphical Memory for Robust Planning

WLPlan: Relational Features for Symbolic Planning

Leveraging LLMs, Graphs and Object Hierarchies for Task Planning in Large-Scale Environments

Trajectory Planning for Autonomous Driving in Unstructured Scenarios Based on Graph Neural Network and Numerical Optimization

Graph Policy Gradients for Large Scale Unlabeled Motion Planning with Constraints

Generalizability of Graph Neural Networks for Decentralized Unlabeled Motion Planning