Manipulation of an object based on dynamic tactile information processing

本研究的目的是开发一种触觉传感方法来识别机器人指尖和物体之间的接触状态，从接触引起的振动，并进一步开发一种控制方法，利用传感器信息。有两个主要结果。(1)基于振动检测的触觉传感器的研制指尖与物体的接触或滑动会在机器人手指的弹性皮肤中产生机械振动。我们已经开发了一种触觉传感器，检测这种振动的压电薄膜（PVDF）嵌入在指尖。瞬态响应和频谱依赖于接触状态的转变。滑移引起的振动是一种宽带的类白噪声信号。滑动速度和接触力对输出的影响较小。检测滑动的开始是在操纵中。代替普通的频域滤波，我们已经开发了一种新的方法，计算波形中的峰值。O处的小振动 关于我们 用这种方法检测了滑移的发生。这种快速响应的滑动传感器是至关重要的稳定灵巧的操作。(2)机器人手的灵巧操作我们已经处理了两指机器人手在桌子上的物体的操作。这项任务涉及灵巧操作中的许多基本运动。关键是区分可控力和被动决定的不可控力。由于真实的世界中固有的不确定性，传统的任务规划方法只依赖于计算可控力是不稳定的。我们已经开发了一种新的方法，以适应接触条件的变化，通过监测传感器的输出，而移动的手指。提起一直握住同一物体的两个手指之一需要调整指尖力的方向。所提出的传感方法使我们能够实现这一任务。灵巧操作中的其他基本运动也是基于同样的概念实现的。整个控制系统能够控制运动，例如连续地在桌子上翻滚物体。少

The purpose of this research is to develop a tactile sensing method to recognize the contact state between the robot finger tip and the object from the vibration induced by the contact, and further develop a control method utilizing the sensor information. There are two major results.(1) Development of Tactile Sensor based on Detecting VibrationContact or slippage between a fingertip and an object generate mechanical vibration in the elastic skin of the robot finger. We have developed a tactile sensor that detects this vibration by piezoelectric film (PVDF) embedded in the fingertip. The transient response and spectrum depend on the contact state transition. The vibration induced by slippage is a broadband white-noise-like signal. Slip velocity and contact force have small effect on the output. Detecting the onset of slippage is in manipulation. Instead of ordinary frequency domain filtering, we have developed a new method that counts peaks in the waveform. The small vibration at the o … More nset of slippage was detected by this method. This quick response of slip sensing was crucial in stable dexterous manipulation.(2) Dexterous Manipulation by Robotic HandWe have dealt with manipulation of an object on a table with two-fingered robotic hand. This task involves many fundamental motions in dexterous manipulation. The key point is to differentiate controllable forces from uncontrollable forces that are passively determined. Conventional task planning method that relies on calculating the controllable forces only is unstable because of the uncertainty inherent in the real world. We have developed a new method to adapt to the changes in contact conditions by monitoring the sensor outputs while moving the fingers. Lifting one of the two fingers that have been holding the same object requires adjustment of direction of the fingertip force. The proposed sensing method enabled us to realize this task. Other basic motions in dexterous manipulation have been realized on the same concept. The overall control system was capable of controlling motions such as tumbling an object on a table continuously. Less