Policies trained on physical test benches

Hardware is only half the problem. By capturing dense tactile pressure arrays and joint telemetry during physical demonstrations, we build robust multi-modal workflows for challenging geometries.

The trained model then runs on your robot in your production.

Four-phase training cycle

Task ingestion

Telemetry collection

Policy tuning

Bench validation

We agree on the part, the robot and what it must do in your production.

We record how the task is done by hand and collect data from the robot and the hand.

We train the robot model on our own GPUs and on the real hand, not only in simulation.

We test the result on a real bench and hand it over to your production.

Tell us your robot and the task. We train it in Daugavpils on the real hardware and hand over a working solution.