Research On Navigation Strategy Of Mobile Robot Based On Deep Reinforcement Learning

Robots have always been the subject of research by scientists.With the increasing pursuit of convenience and efficient production,robots are now playing an increasingly important role.Robots are traditionally stylized.Control has been difficult to meet people's needs.At present,traditional robot navigation technology has insufficient resilience,poor autonomy,and lacks the ability to learn.It is difficult to complete navigation tasks in a changing environment.Therefore,it is especially important for the intelligent research of mobile robot navigation.Aiming at the problem of autonomous navigation of mobile robots in unstructured environment,this paper studies the autonomous navigation strategy of robot based on deep reinforcement learning,which can adapt to the autonomous navigation of mobile robots from random starting point to arbitrary end point,and use deep reinforcement learning to train robots in unknown environment.The navigation behavior allows it to adapt to any unknown unstructured environment,and even if the environment changes,the robot can also perform navigation tasks.The main research work of this paper is as follows:1)Investigate the navigation problems of mobile robots at present,introduce the related technologies to solve the robot navigation problems,and focus on the limitations of the navigation methods of mobile robots at this stage.2)Aiming at the limitation of the mobile robot's own perception of the environment,and the navigation decision problem that the robot has obtained the environmental information and the real-time captured sensory information,a Q-valued deep reinforcement learning algorithm is proposed.(DQN)mobile robot navigation decision method.The DQN algorithm uses the deep convolutional neural network to predict the state of the mobile robot and the corresponding actions,and realizes the end-to-end control of the mobile robot from the environment perception to the decision behavior.The search and utilization balance strategy is used to realize the robot's optimal action.Search,through the construction of the reward function to feedback the advantages and disadvantages of the robot's action,to find the optimal strategy.Finally,the loss function curve of DQN algorithm in the training environment is given.It is proved that the DQN algorithm can get a good convergence effect after a certain training.3)The DQN-based deep reinforcement learning algorithm can not solve thehigh-dimensional continuous action space problem of mobile robots,and the algorithm based on random probability search will lead to the problem that the model is difficult to converge under high-dimensional problems.A mobile robot navigation decision method based on deterministic strategy gradient deep reinforcement learning algorithm(DDPG)is proposed.The DDPG algorithm is based on the Actor-Critic framework and the Deterministic Strategy Gradient(DPG)method,and adaptively updates the algorithm parameters through the evaluation network and the policy network to output deterministic robot navigation behavior.Finally,by comparing and analyzing the loss function and the average q-value curve of DQN algorithm and DDPG algorithm,it is proved that the stability and security of DDPG algorithm are higher.4)Using these two mobile robot navigation algorithms based on deep reinforcement learning to carry out simulation experiments and verify the feasibility of the algorithm.The simulation environment built under the OpenCV platform is used as the experimental platform.The convolutional neural network model created by TensorFlow is used for processing and experiment to realize the navigation of the mobile robot in the simulation environment.The simulation results show that after the deep reinforcement learning method is trained,the mobile robot can still achieve accurate navigation from random starting point to arbitrary end point when some scene changes occur in the environment.