Teach Assistant Experience

2006 Fall, 2007 Spring:
Com S 362: Object-Oriented Analysis and Design

Publications

Yan-Bin Jia, Jiang Tian. "Surface patch reconstruction from minimal tactile data". Submitted to IEEE Transactions on Robotics.

Yan-Bin Jia, Liangchuan Mi, and Jiang Tian. "Surface patch reconstruction via curve sampling" [PDF]. The IEEE International Conference on Robotics and Automation, Orlando, FL, May 15-19, 2006.

Research

1. Tracking Control (From 08/2006 )

The ongoing project is to track along three concurrent curves on the surface. We have recently purchased a BarrettHand from Barrett Technology Inc. We mounted it on our Adept robot. Thus, we have eight DOFs in total. With this new hardware we plan to speed up the data collection on the surface.

BarrettHand (4DOFs)

2. Surface Reconstruction (From 08/2005 )

The ongoing project is reconstructing the surface of the object using tactile data. For a local patch around a point, we use three curves passing through that point to reconstruct the surface. Here, polynomial fitting is used to reconstruct this patch. However, three curves aren't sufficient to constrain the fitting surface. The reconstructed surface has "peaks" and "valleys". Because we assume that there is no drastic change of geometry between the three data curves for a local patch. We would like the reconstructed surface to look smooth. In order to realize this, we minimize "Total Gaussian Curvature" of the reconstructed local patch. Here, "Total Gaussian Curvature" is the integral of the Gaussian curvature over a surface patch. Following is a broken plastic bottle and its reconstructed neck region.

3. Augmented Reality (08/2004 - 07/2005)

The augmented reality system is a low-cost approach consisting only of an IEEE 1394 camera, a mirror, a standard LCD display and a desktop computer. As can be seen in Figure, the LCD is placed in between the user and the desktop. The camera is positioned behind and above the desk in a way that a view of the desktop can be captured in a mirror placed directly behind the LCD. The captured video is then processed in real-time to detect the presence of scene markers using ARToolKIT. 3D virtual graphics are then overlayed on the markers with the correct orientation using OpenGL.