Name the environment variables
Table arrangement, background, lighting, reachable workspace, and object placement can all vary within one location. State which changes the model should handle and which must remain fixed. A new room is not automatically a new manipulation condition, and a rearranged work surface may create meaningful variation without changing the building or camera hardware.
Record initial and target states
An episode needs enough setup context to explain what the robot is asked to change. Identify the source of placement information and whether it is measured or described qualitatively. Preserve the scene before the first task action when required. If resets produce similar initial conditions repeatedly, record that pattern rather than describing every episode as a new layout.
Review coverage across conditions
Check whether tasks, objects, operators, and robot configurations occur across the intended environments. A pooled collection may associate one layout with one task, making the effects difficult to distinguish. Report sparse combinations explicitly. If the project seeks controlled comparisons, describe the planned variation; if it seeks ordinary operation, retain enough metadata to understand what varied naturally.
Choose evaluation groups around the question
Decide whether the test introduces new layouts, new objects, or a new workspace and group related episodes accordingly. Preserve setup identifiers through transformations and additional deliveries. Inspect camera and controller changes at the same time so environment variation is not confused with a different observation or action interface. Source fit depends on these details rather than the location count alone.
Review checklist
- List relevant workspace variables
- Preserve initial and target context
- Review task-condition combinations
- Retain setup identifiers for evaluation