Graph Learning for Numeric Planning
This work addresses the challenge of numeric planning in AI, which extends symbolic planning to include numeric variables, offering incremental improvements in efficiency and generalization for domain-specific applications.
The paper tackled the problem of solving numeric planning tasks by proposing data-efficient and interpretable machine learning models, including a new graph kernel for graphs with continuous and categorical attributes and optimization methods for learning heuristic functions. The results show that their graph kernels are vastly more efficient and generalize better than graph neural networks for numeric planning, with competitive coverage performance compared to domain-independent numeric planners.
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 https://github.com/DillonZChen/goose