01 / INTEGRATION
Device integration
Provide device model, firmware, protocol, timestamp definition, sampling rate and example recordings. We first confirm the data is readable and time-alignable, then pair devices.
DEVELOPERS / INTEGRATION GUIDE
We adapt each project to your devices, data standards and compute environment. This page lists what we need to exchange for integration; ecosystem compatibility goals are not released SDK features.
01 / INTEGRATION
Provide device model, firmware, protocol, timestamp definition, sampling rate and example recordings. We first confirm the data is readable and time-alignable, then pair devices.
02 / INTEGRATION
Define the target model's data fields and action space first, then validate the export format. Video, time-series signals, calibration, annotations and version manifests are delivered together.
03 / INTEGRATION
Agree on inputs and outputs, versions, retries and human review. API keys are managed server-side only; sending sensitive raw data outside requires separate authorization.
04 / INTEGRATION
Define dependencies, model licenses, compute and robot runtime. Before going live, check latency, action limits, emergency stop and human takeover.
START SMALL. THEN INTEGRATE.
Choose one task, one complete episode and the actual device version.
Check clocks, units, coordinate frames, and state and action fields.
Complete reading, playback and export, and record unsupported fields.
Deliver adaptation notes, test results and the ongoing maintenance scope.
OPEN ECOSYSTEM, EXPLICIT SCOPE
Robot-learning toolchains such as LeRobot, ROS 2 device interfaces and multimodal logging tools can be project adaptation targets. Specific devices, data formats and versions are tested first; this does not imply compatibility across all models or any official partnership.
START WITH YOUR TASK
From device selection to data and training workflows, we'll help you find the right place to start.