Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Choosing a Classical Planner with Graph Neural Networks

Jana Vatter, Ruben Mayer, Hans-Arno Jacobsen, Horst Samulowitz, Michael Katz

TL;DR

The paper tackles online planner selection for cost-optimal planning by systematically evaluating four Graph Neural Network architectures on two graph representations (grounded and lifted) and by experimenting with node features and prediction tasks. It further proposes a resource-efficient hybrid approach that uses the GNN-derived graph representation as input to XGBoost, offering strong performance with CPU-friendly training. Across IPC-derived data, grounded graphs and time-based predictions yield the highest accuracy, reaching up to about 0.9 with certain configurations, while feature enhancements like node degree significantly boost performance. The findings provide actionable guidance for building efficient, accurate online planner selectors and suggest several avenues for future improvement, including richer features and specialized architectures tailored to planning graphs.

Abstract

Online planner selection is the task of choosing a solver out of a predefined set for a given planning problem. As planning is computationally hard, the performance of solvers varies greatly on planning problems. Thus, the ability to predict their performance on a given problem is of great importance. While a variety of learning methods have been employed, for classical cost-optimal planning the prevailing approach uses Graph Neural Networks (GNNs). In this work, we continue the line of work on using GNNs for online planner selection. We perform a thorough investigation of the impact of the chosen GNN model, graph representation and node features, as well as prediction task. Going further, we propose using the graph representation obtained by a GNN as an input to the Extreme Gradient Boosting (XGBoost) model, resulting in a more resource-efficient yet accurate approach. We show the effectiveness of a variety of GNN-based online planner selection methods, opening up new exciting avenues for research on online planner selection.

Choosing a Classical Planner with Graph Neural Networks

TL;DR

The paper tackles online planner selection for cost-optimal planning by systematically evaluating four Graph Neural Network architectures on two graph representations (grounded and lifted) and by experimenting with node features and prediction tasks. It further proposes a resource-efficient hybrid approach that uses the GNN-derived graph representation as input to XGBoost, offering strong performance with CPU-friendly training. Across IPC-derived data, grounded graphs and time-based predictions yield the highest accuracy, reaching up to about 0.9 with certain configurations, while feature enhancements like node degree significantly boost performance. The findings provide actionable guidance for building efficient, accurate online planner selectors and suggest several avenues for future improvement, including richer features and specialized architectures tailored to planning graphs.

Abstract

Online planner selection is the task of choosing a solver out of a predefined set for a given planning problem. As planning is computationally hard, the performance of solvers varies greatly on planning problems. Thus, the ability to predict their performance on a given problem is of great importance. While a variety of learning methods have been employed, for classical cost-optimal planning the prevailing approach uses Graph Neural Networks (GNNs). In this work, we continue the line of work on using GNNs for online planner selection. We perform a thorough investigation of the impact of the chosen GNN model, graph representation and node features, as well as prediction task. Going further, we propose using the graph representation obtained by a GNN as an input to the Extreme Gradient Boosting (XGBoost) model, resulting in a more resource-efficient yet accurate approach. We show the effectiveness of a variety of GNN-based online planner selection methods, opening up new exciting avenues for research on online planner selection.
Paper Structure (22 sections, 7 equations, 8 figures)

This paper contains 22 sections, 7 equations, 8 figures.

Table of Contents

  1. Introduction
  2. Related Work
  3. Preliminaries
  4. Graph Neural Networks
  5. Graph Convolutional Network (GCN)
  6. Gated Graph Neural Network (GGNN)
  7. Graph Attention Network (GAT)
  8. Graph Isomporhism Network (GIN)
  9. Extreme Gradient Boosting
  10. Experiments
  11. Methodology
  12. Dataset and features
  13. Tasks and configurations
  14. Training details
  15. Results
  16. ...and 7 more sections

Figures (8)

  • Figure 1: Graph sizes of the grounded and lifted representations
  • Figure 2: Average node degree per node type for the grounded dataset.
  • Figure 3: Results for all tasks and all GNN architectures
  • Figure 4: Feature correlation matrix for the time- and solvable-based task for the grounded and lifted dataset
  • Figure 5: Results for adding the in degree as a node feature
  • ...and 3 more figures