Home

leaf graphic

I am a Ph.D. student in Computer and Information Science at the University of Pennsylvania and a VEF fellow, under the supervision of Dr. Vijay Kumar.  I have been working on deep learning for robotics machine perception. My recent focus is on real-time machine perception on constrained platforms. I received a B.S. degree in Electrical Engineering from Hanoi University of Science and Technology, Hanoi, Vietnam in 2012, followed by an M.S. in Computer Engineering from Ulsan National Institute of Science and Technology, Ulsan, South Korea in 2016. I am also a recipient of the Vietnam Education Foundation Fellowships. 

Contact: tynguyen@seas.upenn.edu

2. Recent Research Projects

1. Fast and robust deep learning models for Visual Inertial Odometry. (April, 2016 – Present)

Visual inertial odometry (VIO) is a technique to estimate the change of a mobile platform in position and orientation overtime using the measurements from on-board cameras and IMU sensor.  VIO has been a highly active research problem due to the miniaturisation in size and low cost in price of two sensing modularities.  However, it is very challenging when accuracy, real-time performance, robustness and operation scale are taken into consideration.  For example, the traditional feature-based approaches to estimating can fail when good features cannot be identified, and can be slow. In this project, we aim to develop deep learning algorithms which can run fast and perform well even when traditional methods fail.

Publications:
1. Ty Nguyen, Steven W. Chen, Shreyas S. Shivakumar, Camillo J. Taylor, and Vijay Kumar. “Unsupervised Deep Homography: A Fast and Robust Homography Estimation Model.” arXiv preprint arXiv:1709.03966 (2017), to appear in ICRA 2018 (pdf)(github).

leaf graphic

Abstract: Homography estimation between multiple aerial images can provide relative pose estimation for collaborative autonomous exploration and monitoring. The usage on a robotic system requires a fast and robust homography estimation algorithm. In this study, we propose an unsupervised learning algorithm that trains a Deep Convolutional Neural Network to estimate planar homographies. We compare the proposed algorithm to traditional feature-based and direct methods, as well as a corresponding supervised learning algorithm. Our empirical results demonstrate that compared to traditional approaches, the unsupervised algorithm achieves faster inference speed, while maintaining comparable or better accuracy and robustness to illumination variation. In addition, on both a synthetic dataset and representative real-world aerial dataset, our unsupervised method has superior adaptability and performance compared to the supervised deep learning method.

leaf graphic

Unsupervised Model Diagram

leaf graphic

 

Advertisements