Improved Hybrid A* Algorithm Obstacle Avoidance Strategy Based on Reinforcement Learning

To address the problem that the traditional obstacle avoidance algorithm does not consider the dynamics constraints and the algorithm planning is not high in real-time, an improved Hybrid A* algorithm based on the reinforcement learning method is proposed. Firstly, the time performance of the Hybrid A* algorithm is improved by remodeling the weight coefficients of the algorithm's heuristic function. Secondly, the algorithm time optimality and performance optimality metric models are established. The corresponding state-action and reward functions are designed incombination with the deep deterministic policy gradient method to train the weight coefficients. Finally, the improved Hybrid A* algorithm is simulated and validated in static and dynamic obstacle environments respectively. The simulation results show effective improvement of the time performance while ensuring optimal path planning results.