| The economy and society of China is under rapid development,the average life expectancy of people has increased rapidly,but the birth rate has been declining,so that the population structure is aging.Nowadays,China has not fully realized modernization,vigorously developing the robot industry is an effective solution to the problem of insufficient labor force under the aging population.Most of the robots currently in use on the market are industrial robots.With the deepening of robot research and the increasing demand for robot functions,domestic and foreign companies and research institutes have launched research and development on service robots.This dissertation designs a home service robot software based on embedded GPU for the application of robot in the family scene,which has functions such as positioning navigation,target recognition and measurement,and human-computer interaction,which has good practical application value.The main work and achievements of the dissertation are as follows:(1)Analyze the function and performance requirements of the home service robot software.The Turtlebot mobile platform and the Nvidia TX2 embedded GPU development board are selected as the hardware development platform.Considering the realization of multi-sensor data acquisition and motion control modules in the robot software,the Robot Operate System ROS is selected as the software development platform.Build a software and hardware development environment,and give the overall design of ROS-based software.(2)Positioning navigation module research and development.By comparing the advantages and disadvantages of the current mainstream SLAM mapping algorithm,the GMapping algorithm suitable for the home environment is selected as the mapping algorithm,and the AMCL algorithm is used to locate the robot.In order to realize robot navigation,A* algorithm is used for global path planning,and DWA algorithm is used for local path planning in case of the Turtlebot's two-wheel differential robot model..In addition,this dissertation also uses deep reinforcement learning to train a neural network,input sensor data,and output control speed of the robot to achieve local path planning control.(3)Target recognition and measurement module research and development.First collected public data sets and crawled the pictures on the network,and produced a data set of commonly used items in the home service scene.Then train different neural network structures,compare the inference speed and accuracy of network,and choose the YOLOv3-Tiny neural network model with better robustness and fast running speed for target detection and classification.After that the identification method of face and humanoid target is designed based on the existing neural network target recognition method.Finally,a depth camera-based target measurement method is introduced.(4)Human-computer interaction module research and development.Firstly,transplanting the Keda Xunfei MSC library to the TX2 development board to realize the function of speech recognition and synthesis.Then the designed humanoid target identification method is used to solve the KCF algorithm tracking drift problem when the target scale changes or the target is occluded,the target tracking function is realized.Finally,based on the designed face recognition method,the multi-person identification function in the home service scene is realized. |
