Tactile texture is a group of vague concepts that describe the sensation of touching a surface. Texture contains vague terms such as roughness, smoothness and softness but can use any term to describe the surface of an object.
This project has two specific aims; to acquire and analyse textures, and to develop an agent that understands texture in human terms. A robotic hand is being developed to acquire electronic textures through active exploration of objects, currently only vibration texture is being recorded, though this will expand to cover other properties of objects as work on the hand progresses. A library of artificial textures is being assembled containing highly constrained set of discs that have specific patterns. These discs will highlight areas that contain information real textures.
Robots that interact with humans will need to be able to understand vague concepts to speed transfer of information and increase their usefulness. Humans use vague concepts to reduce information involved in communication and storage of ideas and experiences, for instance there is no definitive definition of a colour as each person has their own opinions. An agent that can communicate about vague concepts, and even use them to store labels, will find interaction with humans easier than a comparable robot that uses strict definitions. The agents are based around Jonathan Lawry’s work on Label Semantics, and will be tested against other machine learning algorithms like C4.5.
|
