基于改进 PPO 算法的机械臂动态路径规划

Aiming at the increased environmental uncertainties and more difficult modeling for robotic arm path planning in unstructured environments, an approach to dynamic path planning of robotic arms based on an improved PPO algorithm is proposed. In order to solve the problem that the input length of the state space is not fixed due to the change of number of obstacles in dynamic environment, an environmental state input processing method based on the LSTM network is proposed, and the network structure of PPO algorithm is also improved; a reward function is designed based on the artificial potential field method, and a collision detection model of robotic arms is established. Experimental results indicate that the improved PPO algorithm can adapt to the change of the number and position of obstacles in the scene with a faster convergence speed and stability than traditional PPO algorithms.