An unofficial concept study. Enter the passphrase to view.
Most people pick a lane. My work lives in the seams between fields, because that is where the hard, consequential problems hide. The common thread is measurement. If you cannot test it, you cannot trust it.
Defining the tests that turn "seems fine" into evidence a government can procure against.
Making AI safe, explainable, and defensible in the applications where being wrong is expensive.
Intelligent machines that work when it matters: disaster zones, factories, operating rooms.
An intelligent system you cannot secure is a liability, however capable it is.
Turning all of the above into shared rules the field can build and buy against.
Two decades of peer-reviewed work on explainable AI and response robotics, cited more than a thousand times. Live metrics and the full list on Google Scholar.
Explainability is not a feature you bolt on. It is a requirements problem: different applications need different explanations, and the engineering task is matching the need to what the technique can actually deliver.
Standard test methods for the robots that go into rubble, fire, and flood. Developed with DHS and ASTM, hardened at RoboCup Rescue, and pressure-tested in the DARPA Robotics Challenge. If it cannot be scored on a course, it cannot be trusted in a disaster.