Overview
My research spans three distinct areas in vision-based robotics: visual servoing, path planning, and task specification. I am generally interested in developing algorithms that work in unstructured environments. The basic problem that my research addresses is how to specify a task, plan an execution, and accomplish the plan robustly for vision-based robotic platforms in unstructured environments.
My work is on uncalibrated visual servoing, while I also help with the development of our lab's mobile manipulator.
Visual Servoing
The visual servoing problem studies motion control of a robotic manipulator to a desired configuration using visual feedback obtained from one or more cameras. Visual servoing is an interdisciplinary field that combines nonlinear control theory, projective geometry, 3D computer vision, dynamics, and kinematics, and machine learning. There are several approaches to visual servo control. At the University of Alberta, we investigate the uncalibrated image-based visual servoing approaches, where no prior information of the camera or the robot is available. The visual tasks are specified from images and the control law is entirely defined in the image space. During the servoing process, we learn the visual task Jacobian that is used in the nonlinear control law [1]. The visual task Jacobian relates the joint velocity to the rate of change of the visual task, needs to be estimated online.
In unstructured settings, solving the uncalibrated visual servoing is much more challenging due to existence of model outliers. We propose statistically robust methods to detect and reject model outliers. The following poster summeraizes our research results:
