Robotic Experience Learning And Operation Based On Wearable Computer Interaction

Humans as highly intelligent individuals can master new skills and experience in continuous learning.In the continuous learning process,human beings can reserve lots experience to solve the complex problems and tasks.As an industrial product,robots do not have the ability to solve problems and perform tasks at the beginning.To realize robot autonomous task-based operation,robots need a lot of skills through learning.There are two problems that need to be solved in order to master the skills through robot learning.On the one hand,how to obtain the skill-learning with high quality as a learning sample;on the other hand,how to characterize the sample behavior and finally realize the learning of the skill.Learning directly from human experience makes robots more quickly grasp the movement skills.In order to solve the problems above,this paper designs and implements a robot learning and operating system based on wearable devices.The system can be divided into two modules at the design level.The first is a teleoperation system based on wearable device,which is used to obtain high-quality sample of humanoid action based on human experience.The second is a robot learning and operating module based on Dynamic Movement Primitives(DMP)established to understand and learn human activities.The communication mode between the wearable data glove and the robot is designed to realize the real-time transmission of the original trajectory information of the human arm through the data glove to the robot in the teleoperation system.In order to ensure the consistency of motion posture of robot arm and human arm,the method of combining absolute angle,relative angle conversion and direct mapping between human joint and mechanical joint is adopted to realize the mapping transfer from human arm to robot arm.In order to solve the problems of high delay and motion oscillation caused by the original position control,this system uses the manipulator joint motor speed control based on PID control to make the teleoperation system's movement become more compliant and less delay.The system can eventually get high quality motion samples.The robotic experience learning and operating system is built on the establishment of the teleoperation system.In order to understand the movement of the robotic arm after acquired the high-quality movement samples,the designed system uses the improved DMP to realize the feature extraction and learning of the robotic movement.The robot can grasp movement skills through understanding the multi-parameter expression of the robot movements.On the basis of skill learning,we build a motion primitive library for robots to learn and master a set of motion skills of their own.All skills have the properties of rotation invariance,scale invariance and time invariance,so that the robot can implement specific tasks with specific actions when performing different tasks.Finally,the above system is applied to a simple human-computer interaction experiment.Through the demonstration of the experimental results,it can be seen that the simple interaction with mastered skills can be realized and the actions in the interaction are more in line with the human behavior.