| Executing grasp and manipulation tasks in human environments is a necessary function for many robotic applications, and is a research focus in robotic studies. Implementation of such function has raised many challenges to the research of robotic end-effectors.Compared with general grippers and full-actuated robotic hands, the underactuated hands have better grasping performance and some manipulation performance, which gives them research and application value. The number of actuators of an underactuated hand is less than its degree of freedom. This feature enables the underactuated hand to grasp objects of various shapes and conduct power grasp with simple control methods. However, the same feature also generates some difficulties in system modeling, pinch grasp and manipulation, which restricts the development of mechanical design and control method of underactuated hands compared to the requirements for practical applications. Aiming at the challenges of executing complicated manipulation tasks in human environments,this dissertation presents the researches focusing on grasp performance analysis, pinch grasping and manipulation.In order to address the challenges of computation complexity, a grasp performance analysis method based on Permissible Initial Object Position Error(PIOPE) is proposed.The Complementary Problem method is applied to derive the dynamic grasp model of underactuated hand grasping moveable objects and to describe supporting surfaces. Furthermore, a framework to model probabilistic nature of the occurrence of different objects and their initial placement is introduced to define the performance metrics of PIOPE.A grasp performance analysis method is proposed based on the PIOPE to design a prune method with clear physical meanings and to decrease computation complexity of grasping performance analysis. The proposed method is implemented on a 3 phalanx- 2 finger underactuated hand. Considering the cost effectiveness, a reconfigurable palm for changing finger base distance is introduced. Researches on parameter design and structure design are conducted on the underactuated hand.In order to address the challenges of measurement uncertainties for pinch grasping,a grasp strategy based on contact detection and PIOPE is proposed. The method adopts contact detection based grasp control algorithm to address the unexpected movements for objects in the grasping process. Furthermore, the PIOPE is applied to define the objective function for the grasp planner to address measurement uncertainties. In order to address the challenges of uncertainties of the reference model and the noise of signal,Gaussian Process Regression(GPR) is applied to describe the reference model function and the noise function. And thus the accuracy and sensitivity of contact detection are improved. The proposed controller is implemented on an anthropomorphic underactuated i Cub Hand. The experiments results show that the proposed method is capable of improving the performance of pinch grasping comparing to the standard grasp controllers of i Cub Hand.In order to address the model mismatch problem of underactuated hand manipulation model, the enhanced one-step predictive control based on Gaussian process regression is proposed and applied in the study of manipulation. The Gaussian Process Regression(GPR) is applied to model the residuals between the prediction of the principal model and the ground truth data. The compensated model is then applied in one-step predictive control with the rolling optimization implemented by training the Gaussian Process in the on-the-fly fashion to address the model mismatch problem in the manipulation process.The controller is implemented on the i Cub Hand. The experimental results of multiple objects manipulation experiment indicates the proposed method is capable of improving the pinch of manipulation.An underactuated hand platform with two three-phalanx fingers is developed and power grasping, pinch grasping and manipulation experimental research are carried out to verify the proposed grasp performance analysis method, the grasp acquisition and the manipulation method.
|
PDF Full Text Request