Control and Information Process for Multi-Fingers Robot to Acquire Dexterous Grasping and Manipulation

1)稳定抓取和灵巧操作的控制有目标的多指机器人机械系统的动力学约束取决于每个手指接触目标的方式：接触点是固定的或随着滚动或滑动而移动。通过考虑这些因素，从物理角度出发，我们试图从稳定的采摘和灵巧的操作出发，推导出一种新的反馈控制。建立了多指机器人在接触条件变化的情况下抓取物体的动力学模型，在位置和力的混合控制、欠驱动系统的自适应控制等方面获得了大量的知识。2)多指机器人获得灵巧技能的学习机制。证明了动态输出反馈可以镇定非完整哈密顿系统，并通过仿真和实验证明了重复控制即使对非完整哈密顿系统也是有效的。3)基于传感器集成的目标位置和姿态估计方法为了灵活控制多指机器人，需要精确的每个手指与目标接触点的位置信息，从而建立了每个指尖上的力传感器和两个CCD摄像机的传感器集成方法来估计每个手指的接触信息和目标的方向。此外，将软指指尖的应变与力之间的关系建模为某种形式的非线性函数，并通过实验验证了该模型的有效性。

The results obtained during this project are as follows.1) Control for Stable Grasping and Dexterous ManipulationDynamical constraints on the mechanical system of multi-fingers robot with object depend on how each finger touches to the object: the contact point is fixed or moving with rolling or slipping. By taking those into consideration and from physical view, we were trying to derive a new feedback control low far the stable picking and dexterous manipulation. And also we developed a dynamical model for the multi-fingers robot to grasp an object with contact condition changing, then a lot of knowledge have been obtained on hybrid control of position and force, adaptive control for under-actuated systems, etc.2) Leaning Mechanism for Multi-fingers Robot to Acquire Dexterous SkillIt was shown that non-holonomic Hamiltonian systems can be stabilized by dynamic output feedback, and also it was proven by simulations and experiments that the repetitive control is useful even for non-holonomic Hamiltonian systems. Moreover, the conventional repetitive control was modified to work well in the hybrid control of multi-links systems.3) Estimation Method of Object's Position and Orientation by Sensor IntegrationIn order to control the multi-fingers robot dexterously, accurate information on position of each contact point between each finger and object is needed; thus we established a sensor integration method of force sensors on each finger's tip and two CCD cameras to estimate each finger's contact information and the object's orientation. Moreover, a relation between strain and force of soft-finger tip is modeled in a certain form of nonlinear function and it was verified by experiment that the model works very well.