Research And Application Of Deep Q Neural Network Algorithm Combined With Prior Knowledge In Indoor Path Planning

Robot path planning is a popular research direction.Among them,indoor robot path planning task has many problems,such as large indoor environment uncertainty and high safety requirements.However,traditional path planning algorithms,such as global path planning algorithms which need to establish navigation maps according to the environment.It is poor in adaptability to different environments and is not easy to deal with the problem of indoor path planning.Local path planning algorithms tend to fall into local optimum.In order to solve this problem,an indoor path planning model based on deep Q neural network algorithm based on prior knowledge is proposed.The model can complete the path planning task by autonomous learning,and can solve the problems of different indoor scenes better without the need to establish a navigation map,and can obtain the surrounding environment images only through the camera sensor to complete the navigation task.The specific research content of this dissertation is as follows:(1)This dissertation combines the research status of robot path planning and deep reinforcement learning.And proposes a method of deep Q neural network(Priori Knowledge-DQN,PK-DQN)algorithm based on prior knowledge.This method defines and quantifies prior knowledge,and introduces prior knowledge into deep Q neural network algorithm.This method defines a priori knowledge as a rule of action selection,so as to improve the efficiency of the algorithm by reducing the invalid exploration in the training of the algorithm.(2)The path planning model of indoor robot based on PK-DQN algorithm is studied in this dissertation.Combined with indoor robot path planning tasks,in order to ensure the safety of indoor people and robots,a mandatory obstacle avoidance module is set up.The threshold of mandatory obstacle avoidance is set up as a priori knowledge of obstacles that obstruct the safety of robot,and the path planning model of indoor robot based on PK-DQN algorithm is constructed.(3)The dissertation builds the indoor 3D simulation environment based on the ROS robot development platform and the Gazebo simulation software,and designs the simulation experiment of the path planning task of the indoor mobile robot.The experiment is used to verify the effectiveness of the path planning model of the indoor robot based on the PK-DQN algorithm.