We're hiring
Build the human-as-API layer for robots
We make teleoperation software, run a 24x7 robot operations center, and collect research quality training data. We're hiring a Robot Learning Researcher in San Francisco.
Open roles
1 open roleWhat we work on
Aktoria Robotics is building the human-as-API layer for robots. Our teleoperation software runs at state of the art latency, under 8 ms locally and around 30 ms cross country, so a trained operator can act through a robot as if they were standing next to it.
Behind that software is our robot operations center: trained, in-house operators working 24x7 across the globe. And around both, a data practice that produces research quality off-the-shelf datasets plus custom collection on demand, spanning real robot teleop episodes, handheld UMI gripper episodes, stereo and IMU egocentric capture, and custom hardware built to a customer's spec.
The tasks we collect are the awkward, physical ones: cloth folding, water pouring, cable insertion, resin part cleanup, die-cut peeling, loom weaving. If you want to work on robot learning where the data problem and the systems problem sit in the same building, this is that place.
-
Latency is a feature
Sub-8 ms local teleoperation is not a benchmark we quote, it is the standard the product is held to. We care about the milliseconds because operators feel every one of them.
-
Operators in-house
We train and run our own operators rather than outsourcing the hard part. The people doing the work and the people building the tools are on the same team.
-
Data worth publishing on
We hold our datasets to research quality, whether they are off-the-shelf or collected to a customer's spec.
One open role, in San Francisco
We're looking for a Robot Learning Researcher working on VLA and foundation models. If that's you, write to us at [email protected].