This paper proposes a hybrid learning-based navigation framework that integrates classical path planning with a spiking neural network model to achieve scalable and computationally efficient robot navigation in grid - based environments. Optimal reference trajectories are generated using the Spiking Wavefront Planner (SWP) over occupancy maps derived from the KITTI Odometry dataset. These expert paths are employed to train a Super Elman Spiking Neural Network (SE-SNN) using the e-prop learning rule to capture sequential navigation decisions. The framework is evaluated across map sizes ranging from 100×100 to 400×400 and compared with classical planners such as A* and D* Lite. Results show that the proposed SWP-SE-SNN model reduces planning time by up to 94% compared to SWP and approximately 49% relative to A* on large - scale maps, while maintaining comparable or shorter path lengths with a 100% goal-reaching rate. The inference stage exhibits stable runtime behavior as environment size increases, supporting real-time deployment within SDN-based robotic architectures.
