Learning Terrain-Aware Edge Costs for Ant-Colony Route Planning via Genetic Programming

Abstract

Designing transportation routes over complex terrain requires balancing distance with constructability constraints such as slope, elevation, and obstacle avoidance. We propose a hybrid framework that combines Ant Colony Optimization (ACO) with Genetic Programming (GP) on a high-resolution geospatial mesh represented as an 8-neighbor graph enriched with terrain features. GP is used to evolve an edge-cost function via a penalty-driven fitness over multiple start-goal pairs, capturing trade-offs between competing objectives. The learned cost is then exploited by a parallel Min-Max Ant System to efficiently generate feasible routes between user-defined locations. Results show that the proposed method produces computationally efficient routes that reduce steep segments and avoid undesirable areas while maintaining limited detours.