Loading operating surface
Loading operating surface
Industrial Physical AI × World Models · NVIDIA Inception Program Company
Systown puts measured spatial truth into NVIDIA Cosmos 3 world models so humanoids, fleets and people work as one — on the factory floor, in the warehouse, underground and beyond GPS. Deployed on your infrastructure. Proven in simulation before it touches your site.
What we deliver
We simulate your site first — same floor, same fleet, same physics — and only then move a robot. Every number below is labelled with how it was measured.
ThroughputHumanoids, forklifts, AMRs and people re-planned as one every cycle. Potential up to 20% more units per hour — measured in simulation, rounded down.
SafetyEvery machine knows where every person is — indoors, in dust, behind the rack. Potential −20% near-misses between machines and people, measured in simulation.
SovereigntyRuns on-prem or air-gapped on your NVIDIA infrastructure. Deterministic planning proposes; people and guardrails decide. No unverified model output ever moves a robot.
Physical AI × World Models
01 · REAL
Live ±50 cm position and sensor truth, per the Wittra platform specification, from the floors no model can scrape. Powered by Wittra.
02 · SIM
Into NVIDIA Cosmos 3 and our deterministic swarm simulation. Same seeds, same physics — the missing sense on or off.
03 · REAL
swarmCORTEX sends one plan back to the fleet. Potential up to 20% more throughput, fewer near-misses — measured in simulation.
Potential up to 20% swarm-fleet efficiency and safety gains — in simulation, rounded down, labelled on screen. Method, numbers and the live system after your NDA.
The platform
World models, industrial memory and measured position underneath. Systown products and your use cases on top. Every machine and every person reads the same world.
Systown Industrial Physical AI
Product
Reality, queryable — every asset, machine, person and stream in one live world graph, delivered into NVIDIA Cosmos 3.
Product
The swarm, thinking — humanoids, fleets and people re-planned as one, with agentic supervision and an evidence ledger.
Yours
Mission-critical industrial and sovereign scenarios — from line-side co-work to GPS-denied operations — simulated first, then run live.
Brain
Open, omnimodal world models for Physical AI. Trained and optimized on measured reality.
Memory
Decades of trusted industrial data. The plant’s memory becomes the world model’s memory.
Sense
Adaptive narrowband + ultra-wideband 3D positioning and sensing. ±50 cm per the Wittra platform specification, indoors and underground.
Products
Reality, queryable.Every asset, machine, person and sensor stream in one live 3D graph. What is happening, where — one answer.
The swarm, thinking.Humanoids, forklifts and people share one session. Routes cross. Aisles block. The swarm re-plans as one.
Mission-critical · sovereign
World-model-driven autonomy for mission-critical industrial and sovereign operations — where the cloud is not an option, GPS is not available, and every decision must be accountable.
01
The whole stack — world models, spatial graph, fleet cognition — runs inside your perimeter, on your NVIDIA infrastructure. On-prem or air-gapped. Nothing leaves the site.
02
Measured spatial truth, not inference: indoors, underground, between steel, beyond GPS. When cameras go blind, the world model still holds one accurate picture of every machine and every person.
03
Deterministic planning proposes, agentic supervision approves or vetoes, people stay on the loop. Every decision carries its evidence. No unverified model output ever moves a machine.
04
Your site, your fleet, your physics — simulated and measured before anything is deployed. The same loop then runs live: REAL → SIM → REAL, continuously.
Measured results
Five humanoids, five forklifts, ten staff, 1,000 paired seeds. With missionGRAPH + swarmCORTEX on top of NVIDIA Cosmos 3: throughput up, near-misses down. Method, numbers and the live system after your NDA.
Sources: swarmCORTEX V9.2 / V10 evidence (paired-seed simulation, p < 5×10⁻⁵); AVEVA PI data plane commit 280420a. Simulation is always labelled simulated. No field claims. Method, numbers and the live system: after your NDA.
Where it works
One measured world wherever autonomy runs — indoors, underground and beyond GPS.

Defence & securitySample →
Perimeter awareness · patrol robots · GPS-denied corridors

ManufacturingSample →
Line-side AMRs · humanoid co-work · one shared world

Warehouses & logisticsSample →
Swarm dispatch · pallet flow · near-miss prevention

Underground minesSample →
Positioning beyond GPS · vehicle & personnel awareness

Steel & smeltersSample →
Furnace-zone fleets · heat-hardened sensing

Oil & gasSample →
Asset integrity · crew mustering · one site-wide picture

Utilities & energySample →
Substation assets · field crews · energy intelligence

AirportsSample →
Ground ops · equipment flow · apron awareness

Ports & terminalsSample →
Container flow · crane zones · human–machine separation

Data centersSample →
Patrol robots · rack-level assets · climate

Pharma & life sciencesSample →
Cleanroom co-work · fill-finish lines · GMP-grade traceability

HealthcareSample →
Mobile equipment · assistive robots · indoor awareness

Smart cities & municipalitiesSample →
Depot fleets · public assets · works logistics
Questions, answered
An Industrial Physical AI platform: NVIDIA Cosmos 3 world models, the Wittra 4D sense mesh and AVEVA PI industrial data, bound into one operating picture — so humanoids, fleets and people work as one on the factory floor, in the warehouse, underground and beyond GPS.
With a 45-minute briefing on your use case and an NDA. We then simulate your site — same floor, same fleet, same physics — and only after the numbers hold in simulation does anything move on your floor. REAL → SIM → REAL.
Inside your perimeter. The whole stack — world models, spatial graph, fleet cognition — runs on your NVIDIA infrastructure, on-prem or air-gapped. Nothing leaves the site.
Yes — that is the point. The platform is built on measured spatial truth rather than inference: indoors, underground, between steel, in dust and darkness. When cameras go blind, the world model still holds one accurate picture of every machine and every person.
No. Deterministic planning proposes, agentic supervision approves or vetoes, and people stay on the loop. Every decision carries its evidence — no unverified model output ever moves a machine.
Systown is an NVIDIA Inception Program company, and the platform builds on NVIDIA Cosmos 3 world models. Systown is an independent AI company headquartered in Stockholm.
Get in touch
A 45-minute briefing on your use case, an NDA, and a simulation of your floor with your fleet — before anything is deployed. That is how every engagement starts.
Tell us what you are building. Short intro, NDA — then we simulate your site first.
Keynote · AVEVA PI Data Infrastructure Nordics Roadshow · Stockholm · 10 September 2026
Patrik Johansson, Founder & CEO of Systown, keynotes “Physical AI is coming. Is your data ready?” — from industrial data to autonomous operations, and what it takes to make Physical AI work in the real world.
