Bound the task
Specify the items, packaging, environment, handoff, and failure conditions. A narrow task gives us a result we can actually interpret.
Research method
A demo is a hypothesis.
We use simulation to find questions worth testing, then define the physical evidence needed to answer them.
Measure in simulation
Render training data, test perception, and run pick cycles. Report the setup and the limits alongside each result.
Study the failures
Track missed detections, bad grasps, drops, and recovery. A success rate without a failure breakdown hides the next engineering problem.
Validate outside simulation
Before deployment, test physical packages, lighting changes, safety, intervention frequency, maintenance, and installed economics with an operator.
What we do not claim.
A synthetic-data score is not a real-camera score. A rendered cycle is not a throughput guarantee. A cost model is not a supplier quote. Retail Lab remains simulation-stage research; no physical store has been deployed by this lab.