AI-Guided A-Star Path Planning with Predictive Search Corridors and Adaptive Expansion

Abstract

This paper investigates reducing unnecessary node expansion in A* path planning through machine-learning-guided predictive search corridors. While A* guarantees optimal paths with an admissible heuristic, its computational cost increases as graph size and search complexity grow. The proposed method predicts a reduced search corridor before planning begins, allowing A* to focus on high-probability nodes while an adaptive expansion mechanism progressively releases additional candidates when the initial corridor is incomplete. A certification stage compares the resulting path against a global A* search to verify correctness. The approach was evaluated on small- and medium-scale generated geometric graphs using Random Forest, GraphSAGE, and an oracle corridor. On the medium dataset, the GraphSAGE-based method reduced node expansions compared with standard A* in 23.6% of cases, matched A* in 69.9%, and remained uncertified in 6.5% of cases. The oracle corridor improved over A* in 55.1% of cases while matching the remainder, indicating that further reductions in search effort are theoretically achievable. These results demonstrate that predictive search corridors combined with adaptive expansion provide a promising direction for improving the efficiency of optimal path planning while preserving correctness.