USING TACTILE DATA FOR REAL-TIME FEEDBACK

Object recognition through the use of input from multiple sensors is an important aspect of an autonomous manipulation system. In tactile object recognition, it is necessary to determine the location and orientation of object edges and surfaces. We propose a controller that utilizes a tactile sensor in the feedback loop of a manipulator to track edges in real time. In our control system, the data from the tactile sensor is processed in two stages to determine the location of edges. The first stage involves adaptive thresholding of tactile images. This thresholded image is used to detect edges. In the second stage, we use a combination of weighted least squares and the Modified Adaptive Hough Transform to choose the edge to be tracked from the multiple edges present in the tactile image. The parameters of these edges are then used to generate a set-point signal to drive the manipulator. In our implementation, we use a Hybrid controller that uses both position and force set points. The edge tracker, presented in this article, has been implemented on the CMU DD Arm II system. While the signal processing and the control computation require 30 ms, the tactile images are transferred from the sensor to the computer every 240 ms. Consequently, the acquisition of tactile images limits the sampling rate to 4.2 Hz. In this article, we describe both the theory and experimental implementation of tactile edge detection and an edge tracking controller.