Why I’m combining VR, AI, and robotics

In 2013 I put on an Oculus DK1, loaded the Tuscany demo, and stood in a villa that didn’t exist. I was completely awestruck. Over a decade of professional VR work later, that feeling hasn’t worn off — but something about the landscape has changed, and it’s the reason this channel exists.

Three fields, one desk

VR, AI, and robotics used to be three separate labs. A decent motion-capture setup, a machine-learning cluster, and a robotics workshop were each institutional-budget purchases. In 2026, the equivalent kit is: a Quest 3, a gaming PC, and a box of parts from any electronics shop. Call it two months’ rent, total.

That’s not just cheaper — it changes who gets to do this. The convergence of these fields stopped being a research-paper abstraction and became something you can literally do at the kitchen table. I find that mildly absurd and completely irresistible.

Why the overlap matters more than the parts

Each field on its own is well covered. There are excellent VR channels, excellent ML channels, excellent electronics channels. The gap is in the seams:

• VR + AI: agents you can stand next to. Training an ML-Agents character is interesting on a monitor — it’s something else entirely when you share a room with it in a headset.
• AI + robotics: sim-to-real. Train a behaviour in Unity at a thousand times real speed, then move it onto motors and wheels and watch reality file its objections.
• VR + robotics: headsets are sensor platforms. The Quest 3 robot exists because a headset is, frankly, an outrageous amount of robot for the money.

The projects on this site all live in those seams. That’s deliberate. It’s where the surprising results are — and where the least has been written about what actually works.

What “ludic” means here

The name comes from ludus — Latin for play, game, sandbox. Not because the work isn’t serious, but because play is honestly the best research method available to one person with a desk. Try things, break things, follow what’s interesting. The channel documents that process — including the parts that don’t work, because those tend to be the most useful videos to find at 2am.

Where this is heading

Short term: more tutorials (the ML-Agents series is ongoing), more robot iterations, and the written companions you’re reading now. Medium term: structured courses for people who want a guided path rather than a trail of videos. Long term — well, that’s what the experiments are for.

If any of that sounds like your kind of sandbox, subscribe on YouTube, or stick around and read the field notes. Let’s build something.