OUR APPROACH

Built on robotics.
Designed for people.

You know your task and what a good result looks like. We’re building robomato to help small teams use that knowledge with a robot, without assembling the robotics software themselves.

A working skill before a training session

Our intended first experience starts with a supplied skill for a supported robot. A skill is more than model weights: it needs a known sensor setup, calibration checks, operating conditions, and a way to evaluate the result.

A conversation with tools behind it

The assistant should inspect the setup, explain what is possible, ask for missing information, and guide the next step. Understanding a request does not mean a robot can perform it. The interface must keep that distinction clear.

Teach, test, then decide

Guided demonstrations should help you adapt a skill. Training and evaluation should be visible, and a candidate should not reach hardware just because its training score improved.

Start with a defined robot setup

Our first validation work uses supervised tabletop tasks with known objects. The product direction is skills for existing equipment. Each supported setup needs an agreed robot, end effector, sensors, equipment configuration, calibration, and operating limits.

Technical options under evaluation

We are exploring ROS 2 integration, MuJoCo simulation, and existing model runtimes. Reusable components, including OpenRAL, could provide parts of the execution layer.

Transport choices such as DDS or Zenoh are still to be evaluated. Lichtblick is an option for visualization and diagnostics, not a transport replacement. These are engineering options, not features a user should need to configure or promises of shipping support.

Software-in-the-loop and hardware-in-the-loop are different test arrangements. A simulation result is evidence about that simulation—not proof that a physical deployment is safe or reliable.

Keep authority and safety explicit

The planned experience requires authorization before physical execution. Robot-side limits and stopping behaviour must remain independent of the assistant. A conversation is not a safety controller.

Documentation and access

Public installation guides and compatibility lists are not ready yet. We will publish them with a release that has a clear support boundary. For now, read the development status or find the team on GitHub.

Explore the experience