RBE PhD Speaking Qualifiers
Albert Enyedy
Stereo Image-Based Visual Servoing Towards Feature-based Grasping
Friday, May 5th, 2023
8:30 AM - 10:00 AM
Unity Hall Rm 150E
Abstract: In this paper we present an image-based visual servoing scheme that can control robotic manipulators in 3D space without performing stereo reconstruction of the acquired images. We use a stereo camera in an eye-to-hand configuration to reach target positions by directly mapping image space error to joint space actuation, without using camera calibration, 3D vision training datasets, or motion planning. By doing so, we avoid calibration errors and reduce the steps required to reach a goal. We conduct several experiments in simulation and on two different physical robots consisting of different tasks: reaching a static target, reaching a target from an unknown configuration, and grasping the handle of an object. Our results demonstrate that this controller achieves convergence to a static target, even after walking to a new location (for the Atlas robot), and grasps the handle of a practical object, all via only 2D features. We believe that this algorithm is a significant step towards enabling intuitive interfaces that enable a user to initiate a grasp on an object by specifying a grasping point.
Advisor:
Professor Berk Calli, Worcester Polytechnic Institute (WPI)
Committee:
Professor Farzan Siavash, Worcester Polytechnic Institute (WPI)
Professor Jane Li, Worcester Polytechnic Institute (WPI)