<p>Generalized planning using deep reinforcement learning (RL) combined with graph neural networks (GNNs) has shown promising results in various symbolic planning domains described by planning domain definition language (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 grid-based environments. This dense representation results in diluted node-level information, exponentially increases memory requirements, and ultimately makes learning infeasible for grid-based problems. To address these challenges, we propose a sparse, goal-aware GNN representation that selectively encodes locally relevant relationships and goal-relative spatial features, thereby improving computational efficiency. In addition, we employ curriculum learning to progressively scale task difficulty, which enhances training stability in larger grid-based environments. 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 makes RL-based generalized planning practical in grid-based PDDL environments that were previously infeasible with dense graph representations, while also improving training stability, policy generalization, and overall success rates.</p>