Sensing strategy
Mounting, exposure, and sync that still work after an assembly line change.
Hardware Form / Physical AI
Sense · infer · act · manufacture
Physical AI
Machines that read the room, choose an action, and move with limits you can defend in production.
Physical AI is the combination of sensing, edge inference, actuation, and closed-loop test on hardware you can build repeatedly. Hardware Form owns architecture through SMT, assembly, and NPI so model updates stay tied to BOM revision and factory test.
How we scope a program
| Layer | Engineering focus | Manufacturing focus |
|---|---|---|
| Sense | Camera, IMU, force, lidar selection | Calibration fixtures, cable strain relief |
| Infer | SoM, thermal, power budgets | PCBA test, model OTA pinning |
| Act | Drives, brakes, soft limits | Torque tools, functional safety hooks |
| Prove | HIL, sim correlation | EOL sequences shared with EVT |
Edge inference
Modules and thermal paths chosen with factory test in mind.
Actuation
Mechanisms and control limits matched to real payloads.
Closed-loop testing
HIL racks and line tests that share the same scripts.
Physical AI FAQ
What is physical AI in product terms?
Embodied systems that perceive, decide locally, and act on the world, with hardware and manufacturing treated as one program.
Do you train foundation models?
No. We build devices, capture rigs, and production lines; your ML team or partners own training.
Where does data collection fit?
When you need demonstrations or field logs, we build capture hardware aligned with the robots you ship. See robotics training data.