Robotic Arm Pile Sorting
Date Submitted
Fall 12-12-2016
Research Mentor and Department
Dr. Arye Nehorai, Electrical & Systems Engineering Department
Restricted/Unrestricted
Unrestricted
Abstract
The objective of this project was to have a 6-axis FANUC robotic arm, in conjunction with a stationary 3D Kinect camera, pick up, then sort objects that were in a randomly oriented pile. In order to achieve this, we developed image processing algorithms in Matlab that analyzed images from the Kinect. These computer vision algorithms employed: Canny edge detection, Hough Transforms, image combination, and image filtering, in order to search for a location where the robotic arm would have a high chance of success to pick up a target object. After identifying this location in the Kinect image, we performed a coordinate transformation to translate these coordinates (XYZ and Roll-Pitch-Yaw), from the Kinect’s coordinate system to the FANUC’s coordinate system. This coordinate transform was performed using a calibration method based on 64 sample points. Once a target object was picked up, it could then be moved to a fixed position where it could be easily identified and sorted. Alternate methods such as 3D point cloud model detection, point feature matching, and object training were also tested, but proved to be infeasible.