| The development of science and technology has led to the widespread use of mobile robots in various production and daily activities,which has greatly benefited humans.The problem of path planning is a research hotspot in mobile robot technology.Focused on this issue,the Turtle Bot3 Burger mobile robot was used as the research subject,and to complete the path planning task of mobile robots in dynamic environments was token as the research goal.Based on the Twin Delayed Deep Deterministic Policy Gradient(TD3)algorithm,path planning algorithms of mobile robots in dynamic environments was studied.A path planning algorithm based on the improved TD3 algorithm was designed to address the problems encountered in the task,and the comparative research results demonstrated that the performance of the path planning was significantly improved.In the thesis,the Turtlebot3 Burger mobile robot was firstly established in the ROS system environment by writing model files.Based on its characteristics and experimental requirements,an environment file was written to implement a visualized robot model and dynamic simulation experimental environment in Gazebo 9.On this basis,the kinematic model of the mobile robot was established and analyzed,and the state space,action space,and reward function of the robot in the experiment were determined based on the principles of the laser radar and dynamic environmental parameters.A path planning algorithm framework for mobile robots in dynamic environments based on the TD3 algorithm was designed,and a simulation experiment was conducted.In the simulation experiment,the algorithm achieved the path planning task,but the convergence time was too long during training,and the success rate was not high in the testing phase.Then,the above causes of the TD3 algorithm in dynamic environment path planning tasks was analyzed from four aspects,and improvement methods was designed for each of them.Specifically,the methods included prioritized experience replay to address the impact of failed experiences on the training process,transfer learning to obtain suitable initial weight parameters in the current training for the next training to shorten the training time,introducing OU noise to optimize the robot’s exploration method during training,and designing dynamic delayed update strategy to make the update step more reasonable and effectively avoid the influence of local minimum values.These improvement methods were integrated into the TD3 algorithm and first applied to simple static environment path planning tasks.Through comparative experiments,the effectiveness of the improved TD3 algorithm was preliminarily verified.Then,the improved TD3 algorithm was applied to the path planning task of mobile robots in dynamic environment.Comparative experimental results showed that the test success rate of the improved TD3 algorithm relative to the DDPG(Deep Deterministic Policy Gradient)algorithm was increased by 22.4% in the same dynamic simulation environment designed in this thesis,and the total training time was approximately shortened by 2h.Compared with the TD3 algorithm,the success rate of test was improved by 16.6%,and the training time was approximately shortened by 3h,verifying the performance improvement of the path planning algorithm based on the improved TD3 algorithm in dynamic environment.Finally,as the improved TD3 algorithm was tested in a specific simulated environment,in order to verify its adaptability to different environments,two progressively complex dynamic environments were established.And a generalization experiment was designed that included independent training and testing phases.The experimental training achieved good convergence,and the test results showed that the success rates of the mobile robot reaching the target points were 91.3% and 85.5%,respectively.This demonstrates that the improved TD3 algorithm has strong generalization ability in solving the path planning problem of mobile robots in dynamic environments. |
PDF Full Text Request