Adaptive A*–Q-Learning–DWA Fusion with Dynamic Heuristic Adjustment for Safe Path Planning in Spraying Robots

摘要

In underground coal mines, narrow and irregular tunnels, dust, and gas hazards pose significant challenges to robotic path planning, particularly for shotcrete operations. The traditional A* algorithm has the limitations of limited safety, excessive node expansion, and insufficient dynamic obstacle avoidance capabilities. To address these, a hybrid algorithm integrating adaptive A*, Q-learning, and the Dynamic Window Approach (DWA) is proposed. The A* component is enhanced through improvements to its evaluation function and node selection strategy, incorporating dynamically adjustable neighborhood sampling to improve search efficiency. Q-learning re-plans unsafe trajectories in complex environments using a redesigned reward mechanism and an adaptive exploration strategy. The DWA module facilitates real-time obstacle avoidance in dynamic scenarios by optimizing both the velocity space and the trajectory evaluation process. The simulation results indicate that the proposed algorithm reduces the number of path nodes by approximately 30%, reduces the computational time by approximately 20% on 200 × 200 grids, and increases the path length by only 10%. These results demonstrate that the proposed approach effectively balances global path optimality with local adaptability, significantly improving the safety and real-time performance in complex underground environments.