Matthew Johnson-Roberson: Beyond the dataset: addressing training, deployability, and scaling in deep learning-based perception for field robots

April 5, 2017 @ 12:00 pm – 1:00 pm
B17 Hackerman Hall


Robotic platforms now deliver vast amounts of sensor data from large unstructured environments. In attempting to process and interpret this data there are many unique challenges in bridging the gap between prerecorded datasets and the field. This talk will present recent work addressing the application of deep learning techniques to robotic perception. Deep learning has pushed successes in many computer vision tasks through the use of standardized datasets. We focus on solutions to several novel problems that arise when attempting to deploy such techniques on fielded robotic systems. The themes of the talk are twofold:  1) How can we integrate such learning techniques into the traditional probabilistic tools that are well known in robotics? and 2) Are there ways of avoiding the labor-intensive human labeling required for supervised learning? These questions give rise to several lines of research based around dimensionality reduction, adversarial learning, and simulation. We will show this work applied to three domains: self-driving cars, acoustic localization, and optical underwater reconstruction. This talk will show results on field data from the monitoring of Australia’s Coral Reefs, the archeological mapping of a 5,000-year-old submerged city, and the operation of a level-4 self-driving car in urban environments.


Matthew Johnson-Roberson is Assistant Professor of Engineering in the Department of Naval Architecture & Marine Engineering and the Department of Electrical Engineering and Computer Science at the University of Michigan. He received a PhD from the University of Sydney in 2010. There he worked on Autonomous Underwater Vehicles for long-term environment monitoring. Upon joining the University of Michigan faculty in 2013, he created the DROP (Deep Robot Optical Perception) Lab, which researches a wide variety of perception problems in robotics including SLAM, 3D reconstruction, scene understanding, data mining, and visualization. He has held prior postdoctoral appointments with the Centre for Autonomous Systems – CAS at KTH Royal Institute of Technology in Stockholm and the Australian Centre for Field Robotics at the University of Sydney. He is a recipient of the NSF CAREER award (2015).

Laboratory for Computational Sensing + Robotics