The body gets a brain, and the boring robot gets the customer

A foundation model that plans and a controller that executes is settling as the field's default architecture. Meanwhile the company with actual warehouse work dropped the legs.

Gemini Robotics ER 2 is positioned as the high-level brain — spatial reasoning, multi-step planning, coordination between different robots — sitting above the control stack rather than inside it.

Why the split is the right one

Low-level control needs deterministic latency and hard safety guarantees that a large model cannot provide. Planning needs world knowledge and language grounding a controller has no way to acquire. Putting the foundation model above the control loop resolves that conflict rather than trading one failure mode for another.

Coordination across heterogeneous hardware is the more forward-looking claim, because it makes the planner the integration point for a mixed fleet — which is what real facilities actually operate, as opposed to the single-vendor deployments in the demos.

And it inherits the multi-agent problem

Several robots coordinated by a shared planner is precisely the interaction surface DeepMind's own multi-agent safety call describes as under-researched, with the added property that failures here have mass and momentum.

The unglamorous bet

Avatar Robotics raised $6.5 million and is in post-pilot expansion with a multi-billion-dollar warehouse operator on sorting and picking.

Semi-humanoid is the commercially interesting word. Full humanoids carry the hardest problems in the field — dynamic balance, whole-body control, cost — for capabilities most warehouse tasks never use. Dropping bipedal locomotion drops the cost curve that the sector's own analysts name as the thing preventing humanoids scaling on factory floors.

Against $23 billion of physical AI funding this year, a $6.5 million seed attached to a real post-pilot expansion is the more legible signal than a $152 million Series A at a $1.35 billion valuation. Pilots are cheap and abundant. The transition to expansion is where most robotics deployments quietly end.

Google — Introducing Gemini Robotics ER 2 → · The Robot Report — Avatar Robotics raises $6.5M seed round to address industrial labor constraints → · The Robot Report — Humanoids won't scale on factory floors until costs drop →