Research On Key Technologies Of Virtual Ultrasonic Robot Based On Digital Twin And Deep Reinforcement Learning

Ultrasound scanning is widely used in the medical field due to its non-invasive,convenient,and fast advantages.As people pay more attention to their health,the demand for ultrasound scanning is also increasing.However,for doctors,ultrasound scanning requires long-term hand control,which leads to a high incidence of hand muscles and an increased risk of occupational diseases such as frozen shoulder;at the same time,in terms of ultrasound scanning itself,the scanned images The quality is affected by the doctor’s operating experience,and there may be a shortage of experienced doctors in some remote areas.The clinical application of ultrasonography has been greatly limited,so ultrasonic robots have emerged to solve the above problems.However,on the one hand,for some complex scenarios,ultrasonic robots cannot completely replace doctors,and still require doctors to work or human-machine collaboration;on the other hand,ultrasonic robot scanning requires appropriate forceposition control,and its operating trajectory needs to be manually adjusted for different scenarios,unable to cope with environmental changes.The existing solution is to obtain a rough operation trajectory based on visual servoing,and then use a complex forceposition control algorithm to train the ultrasonic robot to obtain the operation trajectory.However,this method requires repeated iterative training of the robot during training,resulting in Joints and motors are worn out,reducing the service life of the ultrasonic robot.In order to solve the above problems,this thesis proposes to use the method of digital twin to imitate the ultrasonic robot,and build a set of ultrasonic robot operation and training system based on digital twin and deep reinforcement learning.The specific work content and contributions are as follows:(1)It is proposed to combine the latest digital twin five-dimensional model to build a virtual ultrasonic robot system.The establishment of the model is completed by software such as Blender and Unity.The virtual ultrasonic robot of the digital twin is exactly the same as the real robot,and a data communication module is established between the two to obtain the real-time information of the ultrasonic robot scanning operation can more accurately reflect the real operation situation,complete efficient interaction,and lay the foundation for the following training research.(2)It is proposed to use the digital twin virtual ultrasonic robot to train the medical staff.The medical staff are trained in the virtual scene.After the training,they can adapt to the work scene faster in real work,avoiding accidents caused by direct operation of the robot,which can not only reduce physical training.The cost can make medical staff more familiar with the use of ultrasonic robots,accumulate ultrasonic scanning experience in virtual scenes,and solve the problem of insufficient operating experience of medical staff when operating semi-autonomous ultrasonic robots.(3)It is proposed to use deep reinforcement learning to train the virtual ultrasonic robot in the digital twin environment,and finally map the training results to the real ultrasonic robot directly,so as to solve the problem of repetitive iterative calculation cost of the motion trajectory during the optimization process of the autonomous ultrasonic robot’s operation force position control.High cost,high time cost,serious hardware loss during ultrasonic robot training,reduced operation accuracy and reduced service life,to improve the deficiencies of traditional methods.(4)A virtual ultrasonic robot operation and training system based on digital twins and deep reinforcement learning was built.The system can not only realize the remote operation,but also realize the autonomous scanning operation of the ultrasonic robot.At the same time,the system also integrates the virtual training function mentioned above,integrating operation and training functions.In this thesis,the system is built based on the existing software and hardware platforms,and the physical system is tested,and semi-autonomous and autonomous ultrasonic scanning operations are simulated for experimental verification.In this thesis,the function test of the established system is carried out,and the virtual training experiment is carried out for the human-machine collaborative operation and the ultrasonic robot autonomous scanning operation.After completing the virtual training,use the real robot for job verification.Experiments show that this system can complete the tasks of ultrasonic robot monitoring and training,and the data obtained from the operation training in the virtual environment is effective for the real robot operation,which greatly reduces the training cost of man-machine collaboration and robot autonomous operation,and improves the operation efficiency.