Research On Indoor Target Path Planning Based On Deep Reinforcement Learning

In recent years,artificial intelligence technology has attracted the attention of many scholars and continues to maintain its popularity.As an important branch of artificial intelligence,the reinforcement learning has been rapidly developed and applied in various fields under the influence of machine learning algorithms.In our daily lives,more and more mobile robots are widely used and serve our work and life.Navigation technology is the basis for daily work of mobile robots,and path planning as the basis of navigation technology has also received attention.Based on the traditional path planning algorithm,when the robot faces an unknown environment,it is difficult to effectively explore a collision-free path from the a point of departure to the destination.We hope that while exploring the path,robots can learn relevant experience,acquire knowledge,continuously learn and accumulate,and improve self-learning ability.Scholars have proposed to complete the path planning task of mobile robots based on the reinforcement learning method,so that the robot can obtain an optimal path through continuous trial and error.Aiming at the path navigation problem of mobile robots in indoor environment,the indoor target path planning based on deep reinforcement learning is studied.The specific research contents are as follows:Firstly,based on the deep deterministic policy gradient(DDPG)algorithm,we propose a PDP-DDPG method which integrates high priority data replay and high similarity data pruning,and processes the sample data sent to network training.It proposes a solution to the problem of first-in-first-out(FIFO)storage methods and random sampling in the replay buffer: selecting high-priority samples to the network for training,while removing similar samples in the buffer and retaining some rare samples.The model of the method is programmed with Tensorflow.The proposed method is verified by the Pendulum experiment which is performed on the OpenAI Gym platform.The experimental results indicate that our method can not only achieve similar performances in a shorter training time,but also accelerate the training process and improve the learning stability and convergence efficiency of the algorithm.Secondly,combined with the proposed PDP-DDPG algorithm,the framework of path planning task using reinforcement learning algorithm is studied and designed.In order to apply reinforcement learning algorithm to robot path planning and achieve good effect of path planning,the mathematical models of robot motion model,environment state space,action state space,reward function and action selection strategy are designed according to the framework,which makes the framework more suitable for research needs,practical application and intelligence.The agent can get expected feedback in time.Finally,the effectiveness of the proposed method and framework is verified by simulation experiments.Based on the robot operating system,first build a simulation experiment software platform of the system framework,introduce the ROS system architecture,create a subscription and release mechanism,and obtain information such as messages and services from the environment.Then,in Gazebo,use Xacro to design the robot structure model and load the environment model.Finally,the intensive learning method is used to carry out 3D simulation experiments,and the experimental results are analyzed and compared.The effectiveness of the path-planning task based on the reinforcement learning method is verified.