Scaling Multi-Agent Epistemic Planning through GNN-Derived Heuristics

2025-08-18

AI Summary

In Multi-Agent Epistemic Planning (MEP), representing belief states as Kripke structures impedes the design of effective traditional heuristics, resulting in poorly guided search and severely limited scalability. Method: This paper introduces Graph Neural Networks (GNNs) to MEP for the first time, directly encoding Kripke models to learn relational patterns from structured belief states. A supervised learning framework is employed to predict state quality and generate data-driven, generalizable heuristic functions. Contribution/Results: The approach overcomes the bottleneck of hand-crafted heuristic design, enabling end-to-end, learned heuristic guidance for search. Experiments on multiple benchmark domains demonstrate substantial improvements in solving efficiency and problem scale handled, validating both effectiveness and scalability of the method in complex multi-agent epistemic planning.

Technology Category

• Multiagent Systems: Multiagent Planning
• Search and Optimization: Heuristic Search
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Multi-agent Epistemic Planning (MEP) is an autonomous planning framework for reasoning about both the physical world and the beliefs of agents, with applications in domains where information flow and awareness among agents are critical. The richness of MEP requires states to be represented as Kripke structures, i.e., directed labeled graphs. This representation limits the applicability of existing heuristics, hindering the scalability of epistemic solvers, which must explore an exponential search space without guidance, resulting often in intractability. To address this, we exploit Graph Neural Networks (GNNs) to learn patterns and relational structures within epistemic states, to guide the planning process. GNNs, which naturally capture the graph-like nature of Kripke models, allow us to derive meaningful estimates of state quality -- e.g., the distance from the nearest goal -- by generalizing knowledge obtained from previously solved planning instances. We integrate these predictive heuristics into an epistemic planning pipeline and evaluate them against standard baselines, showing significant improvements in the scalability of multi-agent epistemic planning.

Problem

• Scaling multi-agent epistemic planning with GNN heuristics
• Overcoming exponential search space in Kripke structures
• Improving state quality estimation via learned relational patterns

Innovation

• GNN-derived heuristics for epistemic planning
• Learning patterns in Kripke structures with GNNs
• Integrating predictive heuristics into planning pipeline