Five years of field-test engineering.

Emergency responders needed a way to state what a response aircraft could actually do, not in a demonstration but in numbers another team could reproduce. I developed, validated, fabricated, and documented aerial test methods and their field apparatus: buildable procedures, machine-readable image targets, and more than ten CAD models.

The CAD was part of the method rather than decoration. It carried enough geometry that another site could build the same setup and obtain comparable results.

Field programs introduce wind, distance, terrain, operator workload, and imperfect observation. I administered tests for a public-safety UAS challenge and evaluated 3D-mapping outputs across six teams; the resulting write-up ran to a 109-slide technical report.

I also ran an outdoor drop-test program for small UAS, with releases from altitudes up to 400 feet measured by non-contact video trajectory analysis. That work is documented in internal NIST technical reporting.

A camera in flight rarely captures the clean, complete view a scoring rule assumes. I designed and implemented an experimental computer-vision scoring workflow in Python that samples robot-camera video, applies partial-target logic, and resolves detections through a versioned score schema. Scoring ran automated, with human review retained where judgment mattered.

The workflow scored more than eight evaluation events, and every score remained traceable to the exact frames behind it. A disputed score could be settled by inspection rather than argument.

International events are where a test method meets other teams' robots. I ran more than one hundred quadruped mobility trials and contributed slip-disc terrain concepts used at RoboCupRescue, ICRA, and IROS.

For the 2024 championship, the arena geometry, including a metric 3D floorplan, was modeled in CAD so another venue could reconstruct it exactly. A test method has to be portable for its scores to mean anything across venues.

A test method starts to matter when it works without its author in the room. I contributed to ASTM E54.09, the standards subcommittee for response robots, through technical review and subcommittee work, including reballot support.

It is slow, careful work, and it is how a lab-proven method becomes something any agency can adopt and run.

Wildfire response broke the assumptions the aerial tests were built on. Visual targets became thermal. Controlled sites became uncontrolled. One capability became three separately testable tasks: identification, inspection, and suppression.

I led the NIST and XPRIZE wildfire drone-response collaboration and co-led the student team that built the new apparatus: the Treefire Post, the Thermal Campfire Omni, and 3D-printed concentric-C thermal acuity targets. The work produced a peer-reviewed conference paper.

The wildfire apparatus team was largely students, and co-leading it was among the most rewarding work of those years.

A test method matters only if other people can pick it up, question it, and improve it. Teaching someone to run a test is also the fastest way to find out whether the procedure is actually written down.