Why Spatial Intelligence, Why Warehouses, Why Now
Physical AI is having a moment. Humanoids that walk, models that fold laundry, vehicles that drive without a person behind the wheel — the demos are striking, and the funding has followed them. But the setting where physical AI is quietest and closest to daily, measurable use isn't the one getting the attention. It's the warehouse.
The shift: from abstract to physical space
For most of the last decade, "AI" meant software that lived on a screen: a model reading text, ranking search results, classifying images drawn from a training set. Physical AI is a different problem. The model has to understand one specific three-dimensional space, not a distribution of examples — a particular floor, a particular aisle, a particular rack that moved four inches since Tuesday.
That distinction changes what's actually hard. The challenge isn't training a bigger model. It's building a live, structured, current representation of a real place — one that updates as fast as the place changes, and holds up even where nobody planned for a sensor to be watching that exact spot.
Why warehouses, why now
Warehouses are the sharpest version of this problem. They are large, dense, and constantly rearranged, with equipment and people moving through them all day. Most of the last decade's warehouse automation spend — warehouse management systems, routing engines, autonomous mobile robots — has gone toward planning and moving product, not toward seeing the floor itself.
That leaves a familiar gap. A system of record says where a pallet should be. A cycle count, taken weekly or monthly, says roughly where it was as of that count. Neither says where it is right now.
The gap mattered less when warehouses changed slowly. It matters more every year, because the pace of change on the floor keeps rising: more SKUs, tighter fulfillment windows, more automation to coordinate around. What's changed recently is that the infrastructure to close the gap is already moving through most large sites, mounted on forklifts and mobile robots that are on the floor regardless.
Complement, not competitor
None of this replaces a WMS or a routing system — those answer a different question: what should happen next. Spatial intelligence answers what is actually happening, right now, on the floor those systems are planning against. A WMS that plans inventory movement is only as reliable as its picture of where inventory currently sits. A routing engine that sequences picks is only as reliable as its picture of which aisles are actually clear.
Staer's position is that these systems are complements, not competitors, and the relationship runs in one direction: the more a site has invested in planning and automation software, the more that investment depends on floor-level truth it probably doesn't have yet.
What spatial intelligence looks like on a live floor
Concretely: cameras already mounted on the forklifts and mobile robots on site build a continuous 3D map of the facility as those vehicles do their normal work. No separate scanning robots, no new hardware fleet to buy, charge, and route around the site.
The map isn't just geometry. It carries meaning — the difference between a rack, a pallet, an empty slot, and an obstacle — and it carries time, tracking what changed since the last pass through that aisle. That's the concrete version of "spatial intelligence": not a point cloud, but a live, structured answer to the questions that used to need someone walking the floor to answer. Is this slot actually empty. Is this aisle actually clear. Did this pallet actually move.
As physical AI moves from demos to daily operations, the sites that get there first won't be the ones with the most robots. They'll be the ones that can see the floor those robots are working on.
Spatial intelligence for autonomous mobile robots.