Platform Development And Algorithm Research Of Mobile Robot Based On Reinforcement Learning

Reinforcement learning is one of the important researches in the field of machine learning,and it has many applications in mobile robot navigation technology.However,current reinforcement learning algorithms have many problems,such as slow convergence speed and poor environmental adaptability,and require high training costs in real applications,which brings many difficulties to the application of mobile robot navigation.Therefore,in response to the above problems,this article mainly completes the following tasks:(1)Build a simulation environment and ROS simulation mobile robot through the Gazebo simulation environment,and build a mobile robot platform in the real environment through the ROS operating system.(2)Aiming at the problem that the current robot navigation strategy needs to spend a lot of time retraining in the environment migration,this paper proposes a robot navigation strategy based on deep reinforcement learning.This strategy uses deep reinforcement learning as the robot's decision-making framework,combined with subsequent features,so that the robot can map the features extracted from the previous environment to the new environment.After adapting to a certain environment,if similar characteristics are encountered in the new environment,the previous strategy will be activated to reduce repetitive training and quickly adapt to the new environment.It is trained and verified in a simulated environment.The experiment shows that the method in this paper can complete the navigation task autonomously,and it can adapt to the new environment faster than the traditional reinforcement learning method.(3)Aiming at the current problem that reinforcement learning requires a lot of training in a real environment to obtain a suitable control strategy,this paper proposes a DDPG algorithm framework based on imitation learning(DDPG-IL).The framework first obtains demonstration data through imitation learning and stores it in the expert library,and at the same time pre-trains the DDPG algorithm.Then,the algorithm reasonably uses the demonstration data and its own exploration data for learning.Finally,when the algorithm reaches the approximate expert level,it gradually becomes ordinary reinforcement learning,and continues to train through selflearning until the algorithm converges to a stable state.The mobile robot platform is built through ROS to conduct experiments in the actual environment.The experimental results show that the algorithm converges faster than ordinary reinforcement learning algorithms and can quickly converge in the real environment.