RESEARCH · WORLD MODELS

RESEARCH

Imagine futures without letting prediction block reflex.

OpenRGD research explores a world-model-agnostic layer for proposing, simulating and evaluating possible futures while the Somatic Runtime remains fast, deterministic and independently safety-gated.

STATUS BOUNDARY

This WAM / World Engine direction is research, not the current normative OpenRGD standard. The existing world_model module is reference material; the frontier architecture remains subject to reconciliation and formalization.

STATE MODEL

Separate what was observed from what is believed or imagined.

01

Observed

Evidence-backed state anchored in perception and history.

02

Belief

The current best interpretation of incomplete or uncertain evidence.

03

Predicted

A possible future generated by a model or simulator.

04

Desired

A target state introduced by goals, tasks or policy.

SPECULATIVE LOOP

Prediction is useful when it remains accountable to evidence.

01Hypothesis
02Imagination
03Evaluation
04Action proposal
05Physical execution
06Evidence

RUNTIME RULE

Prediction must never block reflex.

Heavy world models can operate asynchronously on action chunks and speculative branches. Fast safety and reflexive control stay below them. Confidence, risk and uncertainty determine when more expensive models should be escalated.

BACKENDS

One protocol can host different kinds of predictors.

The research direction is intentionally model-agnostic: deterministic physics, learned video/world models, semantic predictors and manipulation-specific models can all propose future states while preserving provenance and uncertainty.

SOURCE MATERIAL