The physical world doesn't pause while a model makes sense of it. Pallets move, trailers arrive, aisles fill and clear. People and machines make decisions that change what happens next.
For AI to act usefully in that environment, it needs an understanding of space that holds together over time. What is there, where it is, what has changed, and how activity connects.
Staer develops world models grounded in real operations. Warehouses are the starting point, a place where demanding research problems meet direct commercial needs.
From observations to an understanding of a place
A camera captures a view at a particular moment. An operating environment extends beyond that view, and keeps changing after the camera moves on.
A useful world model connects observations across locations and time. An object seen from one vehicle may appear again from another angle. A pallet may move while nobody is looking. A space observed earlier may have changed by the time a machine needs to act.
This raises fundamental questions. What should the model remember? When should new evidence replace an earlier observation? How should it represent uncertainty about something it can no longer see?
These questions are central to Staer's research. They also decide whether a system can be trusted to support a decision on the warehouse floor.
Why warehouses
Warehouses combine structured processes with constant physical change. Plans and records describe how work should proceed. The floor shows what is happening now.
The difference matters. Goods can sit in an unexpected location. A loading area can become congested. A vehicle can spend time searching or waiting without the cause ever reaching the operational systems.
Consider a loading dock. Arrival and departure timestamps establish how long a trailer stayed. Understanding the delay takes more context. When did goods reach the staging area, how did loading progress, and what else was happening nearby?
A record of that activity gives an operator a better basis for investigating the problem. It also poses a deeper challenge for AI, connecting individual observations into a coherent account of a changing environment.
Learning from work already happening
The approach starts with cameras on vehicles already moving through the facility.
As those vehicles carry out their normal work, they capture observations from different positions and at different times. Staer brings those observations together in a shared, continuously updated spatial representation of the site.
That creates a foundation for locating goods, examining movement and understanding activity. It also grounds the world-model research in the conditions the technology ultimately needs to handle. Partial views, changing layouts, occlusions and an environment that keeps moving.
The operational setting provides a demanding test. An understanding of the world has to remain useful as the world changes.
From understanding to action
Today, the platform focuses on making operations visible, with live mapping and capabilities for tracking goods and fleet activity.
The research ambition extends into how that understanding can support prediction, planning and coordination. What might happen if work is assigned differently? How could a change in layout affect movement? What does a mobile robot need to know to act safely and effectively alongside other vehicles and people?
These capabilities require different levels of understanding and validation. Observing an event, explaining it and predicting what follows are distinct challenges, approached step by step.
This is the connection at the heart of Staer. Frontier research shaped by the demands of real operations. Practical problems give concrete ways to test progress, while advances in world models expand what becomes possible.
The work starts in warehouses, with the ambition to build a foundation for machines that can understand and act in a changing physical world.