Vision-based Manipulation Of Deformable Objects

Introducing autonomy of deformable object manipulation can facilitate economically important industries,surgical robots and assistant robots.However,it presents challenges due to the complex properties of deformable objects and the time-varying deformation that is hard to predict and control.Moreover,in practical occasions,it is hard to obtain accurate physical models of deformable objects.Thus,how to develop control methods that are robust to modeling errors and can ensure high accuracy is one of the key challenges.The paper developed vision-based control method to manipulate deformable objects.On the one hand,vision is used to real-timely monitor deformation of objects and movement of manipulators.And visual servoing method is used to control robots to improve operation accuracy.On the other hand,unknown models or parameters are visually estimated online to make the method less dependent on accurate models.For operations that will or will not introduce topology changes of objects,the paper proposes deformation trajectory controller and cutting controller respectively.The proposed vision-based deformation trajectory controller uses uncalibrated camera monitoring deformation of unknown deformable objects.Visual feedback is used to control the motion of manipulators to make feature points in the image plane tracking a desired trajectory,and in the meanwhile,to adaptively estimate unknown deformation model online.The proposed vision-based cutting controller of deformable objects uses calibrated camera monitoring deformation of objects and motion of cutting tool as well as estimate unknown material parameters online.Methods developed to predict deformation and plan cutting path online.Motion and pose of the cutting tool are visually controlled to achieve desired cutting effect in terms of the desired cutting depth and the minimized deformation of the object.Experiments about the deformation and cutting control of deformable objects are designed and implemented.Experiment results validate the feasibility of the proposed methods and the stability of the proposed controllers.