Tactile sensing, perception and shape interpretation (cones)

Dextrous robot hands require intelligent control systems to manipulate unknown objects reliably in a complex environment. Tactile sensing and perception are needed to provide information on contact forces and local shape properties to ensure robust manipulation of objects. This thesis considers aspects of (1) tactile sensing from strain measurements in an elastic medium, (2) estimation of contact force and local shape, and (3) global object inferences from local geometric information at multiple fingers. A cylindrical tactile sensor using capacitive strain measurement techniques has been developed. Orthogonal copper strips separated by a rubber dielectric form an 8 by 12 array of capacitors. Pressure on the sensor surface changes the capacitor plate separation; a simple elastic model predicts this sensor behavior well, and is in agreement with extensive experimental measurements. This model and experimental results are used to examine the relations between sensor depth and sensor spacing, and limits to sensitivity, localization, and aliasing. We consider an unknown second-order surface indenting the finger. Principal curvatures, normal force and location are determined from a 4 by 4 window of strain measurements. Sensor strains are predicted by convolving the spatial impulse response of the rubber skin with the assumed surface pressure distribution derived from a Hertz contact model. Gradient search finds the parameters of the convex second-order shape and the force that best fit the sensor data. Experimental results for cylinders show radius estimation within 10%, orientation estimates within 2.5 degrees, and sub-tactel localization to within 3% of the element spacing. We can infer global properties of an unknown grasped object from local geometric properties, on multiple surface patches in contact. The location and orientation of a particular class of generalized cylinders, i.e. true cones, can be found from three ideal features, such as certain combinations of two edges and a face, or two faces and an edge. A hand control method is proposed that would use this tactile information to manipulate unknown objects.