INTENTION → RELATIONSHIPS → MOTION
One key.
Many possible motions.
Move the keyboard. Change the finger’s pose.
The intention stays with the key.
Drag to orbit · Scroll to zoom · Choose a slide to begin
A PROPOSED EXECUTION LOOP
From intention
to predicted consequence.
BSRL describes what should hold. A world model predicts what an action may cause.
Research architecture. This preview uses a scripted motion generator; no JEPA model, learned controller, or physical robot is connected.
The relationship is the instruction.
Changing the keyboard’s pose changes the resulting motion. The body–key intention stays the same.
NO LEARNED MODEL RUNNING
About this study & research references +
Body Syntonic Robot Language (BSRL) is Todd Muscat’s proposed representation of body–world intention. The syntax shown here is pseudocode. This is an explanatory proof of concept, with a parent Hand frame, articulated MCP, approximate key hinge, and scripted motion. A one-joint geometric calculation keeps the spherical pad on the hinged key while permitting a small slide within the patch. It does not implement contact dynamics, a general robot inverse-kinematics solver, an APT interpreter, a BSRL compiler, or a JEPA model.
The JEPA diagram describes a possible future architecture. An action-conditioned model would need task-specific adaptation and calibrated readouts for pose, timing, and MIDI outcomes. Its latent features do not directly supply those quantities.
V-JEPA 2: prediction and planning ↗ReKep: relational manipulation constraints ↗The APT connection ↗