🤖 AI Summary
Existing deep reinforcement learning (DRL)–based generalized planning methods incorporating graph neural networks (GNNs) suffer from severe scalability bottlenecks in large-scale problems due to fully connected graph representations, which induce combinatorial edge explosion, node feature sparsity, and exponential memory growth.
Method: We propose a goal-aware sparse GNN architecture that explicitly integrates task-level semantic goals with spatial structural features via spatial-goal-guided local relational modeling and adaptive sparse graph construction. Our approach synergistically combines PDDL symbolic priors with grid-based environmental encoding for efficient, scalable state representation.
Contribution/Results: Evaluated on large-scale cooperative UAV mission planning, our method significantly improves policy generalization and task success rates. It is the first to enable end-to-end learned planning in grid environments exceeding 1,000 cells, thereby surpassing the performance limits of conventional dense graph representations.
📝 Abstract
Generalized planning using deep reinforcement learning (RL) combined with graph neural networks (GNNs) has shown promising results in various symbolic planning domains described by PDDL. However, existing approaches typically represent planning states as fully connected graphs, leading to a combinatorial explosion in edge information and substantial sparsity as problem scales grow, especially evident in large grid-based environments. This dense representation results in diluted node-level information, exponentially increases memory requirements, and ultimately makes learning infeasible for larger-scale problems. To address these challenges, we propose a sparse, goal-aware GNN representation that selectively encodes relevant local relationships and explicitly integrates spatial features related to the goal. We validate our approach by designing novel drone mission scenarios based on PDDL within a grid world, effectively simulating realistic mission execution environments. Our experimental results demonstrate that our method scales effectively to larger grid sizes previously infeasible with dense graph representations and substantially improves policy generalization and success rates. Our findings provide a practical foundation for addressing realistic, large-scale generalized planning tasks.