DATA PIPELINE
Robot-learning data: from capture workstation to deliverable datasets
Robot-learning data needs a closed loop of ingest, processing, publish, and evaluation. Workstations enable reproducible experiments; the data platform covers versions, access, and failure recollects.
- Robot-learning data
- Data Engine
- Data Engine
Workstations make one task runnable
A robot-learning workstation combines arms, teleop, vision, and compute around a clear task — from demo logging and baselines to on-robot eval, not just a kit of parts.
Choose Atlas A7 / R1 / D and eHand-6 by workspace; list cameras, end-effectors, fixtures, and safety gear against the task.
The platform keeps data traceable
A typical flow covers ingest and sync, processing and QA, dataset release, training and eval, plus failure recollects. Versions, splits, and export lists should be auditable.
Training compute can be planned separately from on-robot inference and cloud training; align environments and access with sales and engineering.
Related FAQ
- Where does Atlas A7 fit?
- A 7-axis research arm for single-arm studies or as the actuation base for Atlas R1 / D dual-arm workstations — grasping, transport, teleop, and learning eval.
- How is the modular platform configured?
- R1 uses A7 as the actuation base; pick base, lift, vision, teleop, and end-effector by task. Fixed stations, mobile ops, and teleop capture are configurations of one platform.
- Do reference specs equal acceptance results?
- No. Validate logging, calibration, timing, and task results on the delivered configuration; cameras, IPC, end-effectors, and software versions are confirmed per project.
Drawn from public product and solution copy. No customer logos or performance promises. Configuration and acceptance follow the project checklist.