Path Planning For Mobile Platforms In Known Environments Based On Deep Reinforcement Learning

Path planning techniques have long been the most difficult problem in mobile platform control navigation.Reinforcement learning is very suitable for dealing with various situations in mobile platform path planning as a machine learning method with strong self-renewal ability.The path planning method based on deep reinforcement learning is a new method to solve such problems.By implementing the intelligent control of the mobile platform,it can not only greatly improve its autonomy and robustness,but also expand its application range.This paper is based on the application research of deep reinforcement learning.The traditional algorithm design is complicated in the path planning problem,and the original reinforcement learning method is difficult to converge.The framework of DDPG algorithm is applied to the mobile platform and passes through the two-dimensional environment.Simulations in a three-dimensional environment verify the effectiveness and practicability of this scheme.Firstly,the basic theory and related algorithms of reinforcement learning and deep learning are explained and deduced in detail,and the first deep reinforcement learning algorithm is introduced.Then,based on the deficiency of value-based update algorithm,a strategy based on policy update is introduced.By deriving the combination of the two,the algorithm framework used in this paper is derived,and the DDPG algorithm using the framework is deduced in detail.Finally,based on the algorithm framework,an intelligent body with both sensing ability and decision-making ability is designed and trained.The endto-end training model is realized by letting the intelligent body control mobile platform simulate in two-dimensional environment and three-dimensional environment respectively.The validity and practicability of the algorithm are verified.