Deep Reinforcement Learning Based Robot Navigation

Reinforcement learning is the art of trial and error,through continuous experimentation to learn better strategies.In recent years,reinforcement learning has shown strong potential in the fields of games,chess,robots,and other fields.It has become a research hotspot in the field of robotics.The basic idea is to continuously try in the environment through agents to learn better strategies for achieving goals,to achieve the purpose of getting more reward value from the environment.Reinforcement learning algorithm is considered by many people as the key algorithm on the road to general artificial intelligence.In this paper,for the application of reinforcement learning algorithms such as DQN,Double DQN,and Dueling DQN in the specific problem of robot navigation,a series of explorations are made on the design of state space,reward function,action space,network structure and so on.At the same time,on the basis of ROS and Gazebo framework,this paper has tested the open source simulation platform of previous scholars and found its low cross-platform reproducibility,low code efficiency,unstable time step and other problems.A series of improvements have been made to the design of the simulation platform,methods for accelerating simulation and matters needing attention are proposed,and the availability of the simulation platform with only CPU time with limited computing power and the reproducibility of the simulation platform are improved.In the simulation experiment,through the hierarchical characteristics of the navigation robot control system designed in this paper,quantitative tests and experiments,the rationality and effectiveness of the state space,reward function,and action space schemes were verified,and the actual landing on the DQN algorithm was summarized and analyzed.In the parameter selection method,the optimal combination of various parameters of the algorithm is selected,and a very good robot navigation effect and a better model migration ability are obtained.