Compliant end effectors are useful for creating secure grasps in robotic manipulation tasks, because they allow for large static friction and stability resulting from the large contact area between the end effector and the object. The ability to sense the end effectors shape when performing manipulation tasks is also important since it provides an additional sensing modality that is often necessary for fine manipulation. This thesis is concerned with robotic manipulation using end effectors that have (1) a known compliance and (2) the ability to sense their shape when deformed. The experimental testbed motivating this research consists of two three-degree-of-freedom fingers, with shape sensors attached to each fingertip. The shape sensors operate by taking camera images of the inside of a fluid-supported elastic membrane, and using a physical model of the membrane to infer information about the its three-dimensional shape from those images. The manipulator's control system allows the sensor's shape data to be used as feedback in it's motion control algorithms. We employ techniques from computer vision and operations research to optimize the algorithm that converts the sensor's images into shape information, and improve its robustness. We outline a method for calibrating the sensors using a tactile display to deform them by a set of known shapes. We conduct experiments measuring the sensors ability to infer the radii of cylinders that it touches, and present two methods for inferring radii from the shape data. Finally, we propose a framework for precisely describing robotic manipulation tasks, which takes the form of a motion description language designed specifically for robotic manipulation. We implement an interpreter for this language on our experimental testbed.
展开▼
