Research Of Robot Teaching Based On Human-Robot Interaction

In recent years,robotics has developed rapidly.And now robots have the ability to perform different tasks,such as medical robots,industrial robots,and space probes.But the robot itself still has many challenges in dealing with uncertain environments.Current artificial intelligence techniques do not support robots to complete new tasks in an unknown environment without teach programming.In the traditional robot teaching method,the professional is required to use the teach pendant for programming,after which the robot can master the specific work skills.Obviously,this method is too time consuming and inefficient to adapt to tasks that require frequent updates.In order to improve the learning ability of robots,more and more people study and demonstrate learning in the field of robot teaching.Through demonstration learning,the robot can learn the motor skills of a certain task by observing the movement behavior of the human demonstrator.This paper focuses on robot teaching based on human-computer interaction.The main work is:(1)Humanoid robot teaching based on visual interaction and neural network: Constructing a robot teaching system consisting of robot simulation software V-REP and human body capture device Kinect sensor,which allows human demonstrators to visualize body movements to humanoids The robot teaches.Specifically,the system captures human bone information through the Kinect sensor,controls the simulated robot in the V-REP,and records related teaching data.Then,based on the RBF neural network learning algorithm,the teaching data is coded and learned,so that the robot learns skills in multiple teaching training.The experimental results show that the robots in the virtual simulation environment and in the physical world can learn the motion skills of the demonstrators through visual interaction.(2)Robotic teaching based on mixed reality: A hybrid reality-based robot teaching system is built for industrial robotic arms,which allows human demonstrators to teach remote robots and have a sense of presence.Specifically,the LEAP Motion control captures the demonstrator's gesture and palm position information and translates it into position control commands to control the distal robotic arm.The binocular vision captures the scene of the robot and its working environment,and the virtual model of the palm is merged into a mixed reality scene and transmitted to the virtual reality helmet to provide real-time visual feedback for the demonstrator.Here,an extreme learning machine is used to generate a new trajectory from the training data.The experimental results show that the robot teaching system based on mixed reality has a more realistic and natural interactive experience.(3)Research on fatigue degree based on sEMG estimation in robot teaching: In view of the problem that the human demonstrator produces fatigue affecting robot teaching in the process of physical human-computer interaction,a weighted Gaussian mixture model learning algorithm is proposed to compensate the fatigue factor..Firstly,the sEMG signal of the human body is recorded by periodic actions to identify the muscle fatigue phenomenon,and the corresponding fatigue index is proposed to be quantified.Then,in the course of multiple teachings,the fatigue state of the human demonstrator is obtained by tracking changes in the proposed features over time.A weighted Gaussian mixture model algorithm is proposed.The muscle fatigue index is used as the weight of each teaching experiment,and the training data is encoded and learned.The experimental results show that the robot can successfully complete the teaching task,and the generated trajectory is closer to the training trajectory under non-fatigue conditions.