The Platform
Robots that learn from watching, not from code
Robai turns human demonstrations into a deployable motion policy in minutes, without code, an integrator, or weeks of calibration.
The Learning Loop
Demonstration to production policy in minutes
Robai samples video at 30fps, extracts pose keyframes, and splits motion into primitives. Edge training finishes in 10-20 min. No cloud upload or plant data transfer.
- 30fps video capture with pose keyframes
- Motion primitives split tasks into segments
- Edge policy training, 10-20 min per task type
- Hold-out checks before production deployment
Task Success Rate by Demonstration Count
100%
50%
0%
~94%
Demonstrations
On-Device Inference
Local inference, no cloud dependency
Each inference cycle runs on the Robai edge unit beside the controller cabinet. Robot operation continues when the network is down.
| Specification | Value |
|---|---|
| Inference latency (steady state) | Under 12ms per control cycle |
| Model update cycle | Continuous refinement, no line stop required |
| Sensor inputs supported | RGB camera, depth sensor, wrist force-torque |
| Network dependency | None for production inference; cloud sync optional |
Robot Compatibility
Use the controllers already in your plant
Robai uses standard industrial protocols without changing controller firmware.
UR Series (e-Series, CB3)
FANUC R-30iB
KUKA KRC5
ABB IRC5
Yaskawa YRC1000
Doosan M-Series
Not on the list? Contact us about your specific controller.
Start a Pilot
Start a Robai line pilot
30-day pilot, one robot, one task type. No credit card or integrator visit required.