Vision Algorithms for Sensing Soft Robots

Abstract

The aim of this paper is the presentation and verification of computer vision algorithms in order to measure the geometric parameters of soft robots. The materials from which soft robots are made from possess large deformations. Embedded sensors or visual processing algorithms are often used to obtain measurement performance data from these robots. Integration of embedded sensors with soft robots can be cumbersome and expensive, also limiting the performance of a soft actuator. In this paper, implemented visual processing algorithms (thresholding, SAD, SSD and ZNCC) to measure performance data such as angle of motion, degree of bending, radius of curvature in real-time implemented with OpenCV libraries and Webcam is described. Soft RGB colour markers were also produced and firmly glued into the body of the soft robot with no hindrance to movement. Some concepts of visual processing applied include colour tracking, template matching and camera calibration. The execution of vision based motion control to a variety of soft actuators such as bending and wedge-shaped soft actuators was described.