See also my software dashboard and list of references.
ORCID and ImpactStory
I’m a senior research engineer at Applied Physics Laboratory at the University of Washington (my official APL page).
Here are some things I’m into. As a rough sketch, I like building and deploying robots that do interesting and hard thing in the water. I have a background in torpedo-shaped “mapping” robots but my current focus is on Remotely Operated Vehicles (ROVs) (wikipedia) — or whatever ROVs become in the future:
[
]
(http://oceanexplorer.noaa.gov/explorations/06olympic/logs/may26/media/checking.html)
ROPOS ROV and technician (no, that’s not me….)
(click on image to go to original page @ NOAA)
(Image courtesy of Olympic Coast National Marine Sanctuary)
ROVs are attractive for a number of reasons. First, they’re the only way to do up-close-and-personal work in the deep ocean (neglecting the small handful of deep-ocean submersibles), from exploration and sampling for science, to repair, inspection and sampling in the oil patch. There is an existing industry, end user base, and a large number of existing units. This also means many people have thought about what ROVs can, could, or should do; although for many reasons the actual levels of embodied autonomy are low.
Work in the ocean will be increasingly important, and it will continue to be done by ROV-like vehicles. It is inevitable that those vehicles will become more intelligent and autonomous, esp. if it improves their efficiency / utility or lowers operating costs.
How do we make ROVs more “auto”? This can range from simple sensor-based stabilization and navigation to fully autonomous behavior with and without human oversight. It also includes assistive technologies that make existing ROVs more efficient and effective.
(Underwater) 3D Reconstruction and SLAM
For ROVs (and ROV pilots), situational awareness is everything. A small number of video cameras and maybe a sonar must provide all of the information about the ROV’s local environment, for navigation, obstacle avoidance, and manipulation.
This need doesn’t go away as we make ROVs more autonomous. In fact it gets worse, as we replace an intelligent, adaptive, trained operator with a robotic controller. Being able to sense and understand the environment around the ROV is a cornerstone technology for autonomous operation.
For me, this topic also branches out into more general questions of robust life-long SLAM. Despite the successes with SLAM, it’s still typically framed as a mapping technology: a short sortie to produce an accurate 3D model of a space. Our expectations on how well it scales, and its robustness to changes in the world are radically different if we think of it as a localization service which the ROV must rely upon.
Video Analytics, including analysis of data from CamHD
Finally, I’ve been very focused on underwater computer vision geometry, but I quickly discovered a number of people interested in image analysis or understanding. The most common use case was automatic detection or annotation of underwater video. Essentially, it’s getting easier to record video but the costs of watching that video and identifying e.g. fish remain constant. Automatic detection and identification, particularly results which are robust enough to transfer to other vehicles, other cameras, other situations, are something of the holy grail.