The Robotics Data AI Fellowship
A hands-on, fixed-term program for engineers, researchers, and builders to work directly on the data infrastructure behind physical AI, egocentric capture, teleoperation, and simulation pipelines that ship to real robotics teams. It isn't a standard job posting; it's a way to build alongside us on a defined problem, with real ownership.
Apply for a fellowshipNot a job posting, a working arrangement
Fixed-term, not open-ended
Typically 8-12 weeks, scoped around one real problem, a capture pipeline, an annotation tool, a simulation harness, not general busywork.
Paid, not equity-only
Fellows are compensated for the engagement itself. Equity or a full-time offer are possible outcomes, never a substitute for pay.
The outcome isn't fixed
Some fellowships lead to a full-time role. Others turn into an ongoing collaboration, a design partnership, a research partnership, or a standing technical relationship. Neither is assumed going in.
What fellows work on
Egocentric & Teleoperation
Wearable and rig-mounted capture pipelines, teleoperation datasets, and multi-sensor fusion for robot manipulation and navigation.
Simulation & Synthetic Data
Photorealistic simulation environments, domain randomization, and synthetic labeling pipelines that stress-test policies before deployment.
Data Infrastructure & Applied ML
Dataset versioning, quality tooling, and evaluation harnesses that feed directly into robot learning models.
Why work here
- Small team, real problems. You'll work directly on egocentric data collection, teleoperation, and synthetic simulation pipelines that ship to real robotics teams; not just internal prototypes.
- Early and bootstrapped. No layers of process. Decisions move as fast as the work does.
- Direct ownership. What you build is the product, not a slide about the product.
- Part of something bigger.You're helping build the data infrastructure a whole industry, physical AI and robotics, will run on for the next decade.
Who fits well here
- Builder mindset. Comfortable owning ambiguous, unglamorous infrastructure problems end to end.
- Robotics or ML curiosity. Some exposure to computer vision, sensor data, simulation, or applied ML.
- Bias to ship. You'd rather ship working software than write a long design doc about it.
- Comfortable with ambiguity. We're early-stage: roles, scope, and priorities evolve fast.