From camera frames to action streams.
Align the right cameras, sensor modalities, metadata and output schema to your training pipeline. We specify what is captured, what is derived and what is not available.
Specify the streams your model needs.
Modalities are not interchangeable. A file format describes how information is packaged, not whether the recording contains a particular sensor or label.
Single-view colour footage for appearance and task understanding.
Paired camera streams with timing and calibration requirements.
Accelerometer/gyroscope streams for movement context when recorded.
Position and orientation in a defined coordinate system, where available.
Robot state, gripper width or control action logs where supported.
Timestamps, labels, device setup, episode IDs, states and outcomes.
Packaging is part of the scope.
The source recording and buyer training stack determine how we prepare the final output. Support for any one format must be confirmed against actual sample files.
MCAP is a container format for recorded messages; converting MP4 to MCAP does not add IMU, calibration or robot actions that were never captured. LeRobot-style mappings or other specialized formats require individual engineering validation.
Have an exact schema? Send it over.
We can assess the required formats and streams against the available capture or processing workflow.