Robot intelligence is bottlenecked by real-world data. Compute scaled. Models scaled. The physical world did not.

We collect the data that closes this gap.

Manipulation, egocentric, teleoperation, and multi-sensor data at scale. Synchronized video, pose, and tracking streams with layered annotation. Every batch is audited before delivery, with clean and exclusive rights.

We design and build our own capture stack: multi-camera rigs, instrumented grippers, and custom devices for any embodiment. Buy them off the shelf, or co-design with us and go from requirements to production in weeks.

On this data, we are training cross-task transfer models: policies that learn a skill in one domain and carry it to another. In the long run we want robot data and models to have an open home, a Hugging Face for robotics.

Researchers and engineers from CMU, Stanford, Meta, and NVIDIA. Headquartered in San Francisco, collecting worldwide.

If you are training robot foundation models, need data or hardware, or want to join us, write to
sylvia@haptalabs.com