Manipulator Motion Control For Programming By Demonstration In Industrial Assembly

Programming by Demonstration(PbD)is a feasible way to generate the executable code for robots by observing and learning the human's manipulation progress,and then transferring directly the new skills learned to make the robot accomplish the same manipulation task.Such a way can ease the workload of programming for robots and reduce both difficulty and time cost,which is of great significance to the application and popularization of the robot.This thesis considers the PbD issue for industrial assembly.To ensure that the manipulator can complete the work successfully in the PbD system for industrial assembly,this thesis develops the manipulation unit in the system and explores the feedback control issue based on force sense information.The main contents of the research and results of this thesis include:1.Develop the execution unit in the PbD system for industrial assembly and achieve the PbD task combined with visual observation and learning.Based on the ABB IRB120 robot arm,the robot hardware is built according to the requirements of the task scene.Meanwhile,the basic software modules like manipulator-and-platform calibration and motion control of the manipulator were successfully executed.Experiments show that the developed manipulator execution unit can complete the assembly task exactly the same as the human demonstration according to the perceptual data with high accuracy and high reliability.2.Propose a prediction algorithm for execution result of the manipulator based on Support Vector Machine(SVM).The algorithm predicts whether the current task execution is completed or not according to the received force data sequence in the process of task execution.During the offline stage,data of the force sensor mounted at the end of the arm was collected while executing the task,and after preprocess the collected data,the classifier trained by using SVM can reach more than 97.1%Precision.During the online stage,the trained classifier was used to monitor and predict the execution process.Experiments show that the algorithm performs well on predicting the task result as well as controlling the robot arm.3.Propose a force feedback control algorithm for manipulator based on Bayesian Optimization.By using the Gaussian Process on the experiment data to do the data modeling,the algorithm can deal with the problem of existence of observing noise and difficulty to create an accurate model.Then by using the constrained Bayesian optimization algorithm to optimize the control parameters,we balance the contradiction between the search space and the search efficiency in order to obtain the global optimal solution in a shorter period of time.Experiments show that the algorithm does not require accurate modeling of the environment,and can obtain the optimal solution after only a few iterations,thus meeting the efficiency and convergence requirements.