Research

Intelligence emerges from prediction.

Movendi is developing Reality Models — AI systems that learn to predict how the physical world evolves, from observation. The field calls this world models. We arrived from the factory floor, and that changes what we build.

How we got here

For three years we built representations of reality the engineered way: digital twins, computer vision, telemetry, state machines, control systems, language models — assembled into systems that hold up under uncertainty. They work. They also demand more engineering with every deployment.

Can a machine learn these representations directly from experience, rather than through engineered combinations of specialized models and software components?

The Direct Predictive Model

The DPM is our attempt at an answer: self-supervised models that learn the dynamics of physical reality from experience. Not predicting words, images, or pixels — predicting observed reality as it evolves.

We are testing whether object permanence, causality, support, collision, and physical reasoning emerge from prediction alone, without physics engines, symbolic rules, or synthetic simulation.

What is real, and what is not yet

Research is published as it becomes reproducible.

Research Notes are published early — evolving hypotheses, not final conclusions — to invite criticism. MRN-001 introduces Reality Models and the Direct Predictive Model. MRN-002 narrows the question to bounded physical environments and proposes a working definition.