www.jestr.org Obstacle Avoidance Path Optimization of Agricultural Warehouse Handling Robots Based on the Stanley-PPO2 Algorithm

Abstract

The research on obstacle avoidance of agricultural warehouse handling robots is easily disturbed by the motion environment of robots. The path planning of agricultural handling robots is generally subject to unsmooth trajectory, prone to local optimum, and challenged by dynamic obstacle avoidance and tracking performance in path tracking. In this study, a path optimization objective function including kinematic constraints and obstacle potential field models was established, the PPO2 algorithm improved based on the Stanley algorithm was designed, and the obstacle avoidance path optimization model of agricultural warehouse handling robots was solved. Moreover, the limitations of the A* algorithm in generating the global initial path was analyzed, and the initial path was optimized by introducing the Douglas–Peucker path simplification and the B-spline smoothing technique. Next, a mixed control system taking the Stanley algorithm as the basic controller and using the PPO2 network to dynamic adjust parameters was constructed. This system realized the collaborative optimization of dynamic obstacle avoidance and path tracking by designing a complex reward function that integrated path tracking deviation, obstacle distance entropy, and motion stability. The simulation experiment revealed that compared with the traditional PPO2 method, the PPO2 algorithm improved based on the Stanley algorithm performs better in length of path, number of iterations, number of inflection points, stability of optimization, and quality of path, and verified the effectiveness of the improved PPO2 algorithm. The obtained results can efficiently guide agricultural warehouse handling robots to reach the random handling point rapidly, avoid obstacles, satisfy the requirements of handling tasks, and provide theoretical and technical support for the path planning of real agricultural handling robots.