Independent research program

Reality Models

Intelligence emerges from prediction.

The world is not made of words.

Living beings understand the physical world through perception, prediction, and action. We recognize objects, understand motion, anticipate consequences, and continuously adjust β€” without first translating reality into language.

Today’s physical AI systems combine simulation, digital twins, language models, and vision-language-action models to produce remarkable demonstrations. But dependable operation in the real world requires something more: continuous understanding of geometry, motion, causality, uncertainty, and time.

Movendi is developing Reality Models β€” AI systems designed to perceive physical environments, learn their dynamics, predict what happens next, and ultimately act upon them.

MRN-001 Research Note / June 2026

Toward Predictive Models of Reality

Introduces the Direct Predictive Model β€” a class of self-supervised models intended to learn the dynamics of physical reality directly from experience β€” and the hypothesis that intelligence emerges from prediction.

Read the note β†’
MRN-002 Research Note / July 2026

From World Models to Reality Models

Movendi is initiating a foundational research program to define and develop Reality Models: computational systems that learn and maintain a compact, persistent, and predictive internal representation of a bounded physical environment.

Read the note β†’
PAPER NeurIPS 2026 Workshop / September 2026

Position: World Models of Physical Sites Must Discover Their Own State

A world model of one physical site must discover its own state from evidence, not inherit it from its sensors or its designers. We give falsifiable hypotheses and a test, H0, that tells whether a model found a missing cause or only fitted the error.

Accepted as a poster at the NeurIPS 2026 Workshop on World Models in Physical AI.

MRN-003 Research Note / September 2026

Persistence is a correspondence problem

A new scan does not observe the same points again. In many cases, no exact one-to-one correspondence exists. Correspondence should therefore be probabilistic rather than forced.

Read the note β†’
MRN-004 Research Note / October 2026

Persistent Predictive State: A Research Framework for Reality Models

A physical environment persists even when its observations do not. A useful Reality Model must maintain continuity through these changes while remaining able to revise what it believes.

Read the note β†’