Experimental
Stage 5 · Train
Show. Correct. Judge.
Turn domain knowledge into demonstrations, corrections and measurable acceptance criteria.

Domain expertise – not coding – becomes the programming interface.
Teaching workflow
| Stage | Domain expert does | System produces |
|---|---|---|
| Plan | Defines the task, conditions and success | Task specification and evaluation criteria |
| Show | Demonstrates the intended behavior | Structured episode data |
| Review | Inspects replay and outcome | Labels, annotations and failure evidence |
| Correct | Demonstrates better behavior where needed | Corrective examples |
| Judge | Tests against the real task | Accepted result or another iteration |
Named locations
Teach meaningful places without exposing map coordinates to the operator.
Named poses
Create reusable, understandable arm and whole-body states.
Capture
Record movement, task events, annotations and success or failure.
Replay & evaluate
Inspect reproducibility and decide whether behavior is good enough.
Capability honesty
The pipeline is developing. Do not describe it as arbitrary task learning. Each supported workflow needs an explicit task definition, data path, safety envelope and acceptance test.