Bodey
Young Dawgs internship
Virtual Experiences Laboratory · University of Georgia
About me
Épée fencer for 6 years
Senior at East Jackson High
4 years of FTC robotics
Future aerospace engineer
Outline
The lab
Vring
Robot arms
Motion simulator
Fencing tracker
…and how I've been able to do all of it.
The people I work with
Dr. Kyle Johnsen
Lab director
Andrew Rukangu
Graduate researcher
Jordan Smothers
Undergrad mentor
What the lab does
“We build virtual experiences that help people work, learn and heal together.”
The lab's motto
Robots & physical systems
Shared spaces·Health & medicine
Learning by doing
most of my work
Motion simulator
Robot arms platform
Bringing old gear back
Varjo XR-3
Unused for years.
Running again.
The lab was full of equipment nobody had time to make usable.
That became my job.
Vring: my sandbox for an XR ring
Hand tracking gives no feedback. The ring adds capacitive touch, an IMU (motion sensor) and haptics (vibration).
My Unity sandbox for the ring
The Vring, Andrew Rukangu's research project
Study tasks: ring vs. hand tracking
Building the arms platform
4
robot arms on a drivable base
2
of them built by me,
plus parts of the base
Leader moves, follower copies
3D model from Trossen's published arm files
My app: Trossen Control Center
Out of the box, starting the arms meant typing console commands. So I built one place for everything.
1Pick a mode
2Press Run
3Every startup step, checked
4Every arm, live
Written in Python · 238 commits · about 57,000 lines of code
When an arm broke
Controller status
No error
| Joint | Position (rad) | Velocity (rad/s) | Effort (N·m) | Motor temp |
A real motor is never 0 °C
Found it: a loose connector.
Reseated, all 28 joints read live, and my app now flags frozen joints.
Controller read, right leader arm, 29 Sep 2026
Then I let an AI
drive the arms
The task I typed
Put the eraser in the blue outline.
look
AI decides: object 1 is the eraser
pick --object 1
Harness: finds the grip, closes, lifts
place --x 0.42 --y -0.41 --dry-run
AI decides: where it goes
place --x 0.42 --y -0.41
Harness: moves it and sets it down
9 commands, about a minute.
2 decisions were the AI's.
Claude Sonnet 5.5, 12 Nov
A simulator locked to three games
Built in 2010. Three axes: pitch, roll, yaw.
Licensed to one computer.
Three racing games.
No studies, no custom experiences.
My job: get direct control.
Learning the motors' language
Messages captured from the original software
Measured on the rig, 27 Oct–2 Nov
Moving on my own software
Rig homing + jogging16:9 clip, with the operator app on screen
E-stop locks in 9 ms.
Nothing moves unless a person says go.
Measured angles from a real run, 12 Nov
Tuning it with a sensor
Baseline vs. tuned run, 6 Nov
The IMU (inertial measurement unit)
My project: a fencing tracker
A coach can tell you that you hesitated on your lunge.
This would show you the numbers behind it.
500 readings a second
Pose tracking and IMU on video, 16 Nov
How: AI tools
The lab taught me to use AI responsibly: as an accelerant, not a replacement.
330
commits across my 3 projects
→
→
→
Document
notes anyone can pick up
What AI can't do
One AI wrote simulator code where
the emergency stop could be blocked.
Its own tests passed. They had the same mistake built in.
A second review caught it before anything moved.
It also can't fix a loose connector,
or know what the lab actually needs.
Debugging hardware I didn't build
One arm kept faulting. Nothing looked wrong from the outside.
So I opened the controller box.
A connector, crushed by the other cables.
I replaced the cable.
Inside the controller box
The crushed connector
What's different now
| Varjo XR-3 | unused for years | → | running |
| Robot arms | console commands | → | one app, anyone can use |
| AI + arms | hand control only | → | moves objects from one sentence |
| Simulator | locked to three games | → | my own driver |
| Fencing | an idea | → | 500 readings a second |
Thank you. Questions?