Identify the recorded quantity
Joint-space fields describe robot joints, while end-effector fields describe a tool or gripper pose or motion in a stated frame. Either can represent commands or measured state. Ask exactly which quantity is stored and how it was obtained. A field named pose or action is too broad to establish that distinction without its accompanying definition.
Request the necessary context
Joint ordering, units, and robot configuration matter for joint fields. End-effector data needs a reference frame, orientation convention, tool definition, and absolute or relative semantics. If a transformation is proposed, ask what source information supports it. A conversion can depend on robot-specific details that are absent from a generic video or a simplified export. Keep any tool-offset definitions with the corresponding robot configuration.
Review what the target model expects
A learning pipeline may predict one representation while the deployment controller accepts another. Describe that interface explicitly in the sourcing brief. Determine whether gripper control is separate and whether the model needs current state in addition to images. Avoid assuming that an action representation useful for one robot can be applied unchanged to another embodiment.
Inspect representative motion
Review reaching, contact, and orientation changes against the recorded fields. Include a sequence where the distinction between target and achieved motion matters. Preserve original fields when deriving a new representation so interpretation can be audited. The goal of sample review is to expose assumptions and missing information before a project depends on the converted data.
Review checklist
- Identify commands versus measurements
- Specify frames, units, and joint order
- Match the deployment control interface
- Retain source fields behind conversions