Integrated improved ant colony optimization and Q-learning for ice navigation route planning

• A hybrid ACO and Q-learning model is proposed, balancing distance, safety, and ice-breaking cost. • Cosine similarity and dynamic weights enhance the performance of ACO. • ACO-generated paths are used to prune the Q-learning state space. Global warming has accelerated Arctic sea ice retreat and extended navigational windows, creating both opportunities and challenges for Arctic shipping. To enable safe and cost-effective route planning, this study presents an integrated method combining Improved Ant Colony Optimization and Q-Learning (IACO-QL). First, the Ant Colony Optimization (ACO) algorithm is enhanced using global distance guidance, cosine similarity, and a dynamic pheromone evaporation mechanism to improve convergence. The Improved Ant Colony Optimization (IACO) constructs a multi-objective solution space tailored to ice navigation. This solution space is used to prune the state space of Q-Learning (QL) and to initialize the Q-table. Second, QL refines the route using a composite reward function that balances distance, navigational safety, and icebreaking cost. Finally, the Bresenham line algorithm removes redundant nodes while preserving key turning points. Experimental results in representative Arctic scenarios show that IACO-QL improves navigation performance. Compared with traditional methods, it reduces route length by 9.84 % in sparse ice and 6.35 % in dense ice, while the number of turns is reduced by 59.57 % and 62.77 %, respectively. These improvements demonstrate the effectiveness and practical value of the proposed method for intelligent and efficient route planning in polar environments.