The Michigan Engineer News Center

New U-M Aerospace Engineering Professor Vasileios Tzoumas Received Best Paper Award in Robot Vision

New U-M Aerospace Engineering Professor Awarded ICRA Best Paper Award in Robot Vision for research paper written during his career at Massachusetts Institute of Technology (MIT)| Short Read

Vasileios Tzoumas joined U-M Aerospace Engineering faculty this January 2021. One of his recent papers while he was at MIT received the Best Paper Award in Robot Vision from the International Conference on Robotics and Automation (ICRA). Vasileios co-authored the paper with the MIT Leonardo Career Development Assistant Professor Luca Carlone, and the MIT Ph.D. students Heng Yang and Pasquale Antonante. The paper proposed Graduated Non-Convexity (GNC), an algorithm that helps robots to reliably perceive and navigate the world despite deceptive failures (outliers) in the robots’ perception capabilities. Such outliers are typically abundant, caused by incorrect data associations, or even sensor malfunctions, and, if ignored, can brittle the robots’ perception capabilities.  

Enlargeportrait of Professor Vasileios Tzoumas
IMAGE:  Aerospace Engineering Professor Vasileios Tzoumas

GNC, instead, can robustify against outliers a variety of perception capabilities, such as object recognition, scene reconstruction, and simultaneous localization and mapping (SLAM). The algorithm intuitively identifies and discards outlier data, by solving a sequence of surrogate optimization problems of increasing non-convexity (details, including videos, can be found in a recent MathWorks news article). Across benchmarking datasets (on mesh registration, shape alignment, and SLAM), GNC was not only more accurate than the state of the art; GNC was also often faster, requiring fewer computational steps.  

Vasileios now is researching how multiple robots could collaborate using the GNC algorithm. More broadly, he cares for a trustworthy collaborative autonomy of mobile robots, from self-driving cars to drones to satellites, even when those robots operate in challenging conditions such as unmapped areas, cluttered streets, and adversarial conditions.

Vasileios Tzoumas received his Ph.D. in Electrical and Systems Engineering at the University of Pennsylvania (2018). He holds a Master of Arts in Statistics from the Wharton School of Business at the University of Pennsylvania (2016); a Master of Science in Electrical Engineering from the University of Pennsylvania (2016); and a diploma in Electrical and Computer Engineering from the National Technical University of Athens (2012). Vasileios started as an Assistant Professor in the Department of Aerospace Engineering, University of Michigan, Ann Arbor, in January 2021.

portrait of Professor Vasileios Tzoumas
Portrait of Kim Johnson

Contact

Kimberly Johnson
Communications Manager

Aerospace Engineering

(734) 647-4701

3054 FXB

Researchers
A simulation of the landing .

Sticking the landing on Mars: High-powered computing aims to reduce guesswork

As the Mars 2020 launch approaches, a separate effort is using simulations to understand landing dynamics for tomorrow's missions. | Medium Read