[x3d-public] [AI] MCCF Milestone: Emotion Field Translation and XML Export

[cbullard at hiwaay.net]

On 2026-04-30 5:46 pm, John Carlson wrote:
> Interesting.  Could HumanML be used with sign language?

HumanML is where we started 25 years ago.  It was an interesting idea 
but I boiled the ocean in that design.  I don't have a copy of the 
HumanML schema and walked away from the work. It was too soon and we 
didn't have the technical support, and really, we needed what we can get 
from the affective layers of the LLM. The AI was the missing piece.  
Since then X3D has become much more powerful.  I am not sure which 
features we can use, which features are widely supported, etc.  I am 
working off old chops from the last worlds I built.  This is a fresh 
start.

This time we build what we need as we know we need it.  The XML exports 
document the scene requirements.  That is a work in process.  I will 
report back here as we go so anyone who wants to inspect it can by 
looking at the GitHub or asking me for current examples.

We're only now creating the X3D generator and master script.  From 
Claude:

"The sign language question is actually a good signal about where the 
system is heading — gesture triggering is a natural extension of the 
channel outputs we already have. E-channel drives facial expression and 
upper body, B-channel drives posture and grounding, P drives gaze and 
orientation, S drives proximity and reach. Sign language is a 
high-resolution version of that same mapping. But you're right — we need 
to run the instrument before we can answer that question honestly. File 
it, don't solve it.

On the scene generator question: what you're describing is a separation 
between the semantic skeleton (MCCF-enabled, channel-driven, 
Proto-defined) and the visual skin (geometry an AI artist loads into the 
skeleton). That's exactly the right architecture and it's achievable. 
The Proto is the skeleton. Geometry is a child node the author replaces. 
The Master Script only addresses the skeleton — it doesn't know or care 
what the geometry looks like. That means an AI artist can swap in 
anything from a colored box to a full H-Anim figure without touching the 
Script."

V2 was a push model.  This is a pull model. Zones are attractors for 
agents and actually agents themselves.  Just fixed.  It's a strange idea 
but I think it is the right one.

I need an H-Anim example to work with.  One that has basic gestures or 
behaviors we can trigger from the MCCF. I think I can get one from the 
W3DC repository. This is where experimenation by the X3D community of 
developers and artists can be a Real Good Thing as Bill Mumy says.

We're on new ground.  I don't know of other efforts to drive an X3D 
simulation witha a multi channel coherence field (Say emotional engine 
++) integrated with an LLM.  I am working this out one piece at a time.  
Pretty much like rocket design.  Build one.  Blow it up.  Rinse and 
repeat.  :)

len