What should a warehouse robot report after a bad pick?
Warehouse robots can move past millions of easy picks without touching the awkward ones: loose polybags, cloth, porous packages, double picks, damaged goods. Automated Warehouse reports that Locus is preparing a softer gripper for its Array mobile manipulator; its robotics lead says pick quality and reliability matter more than speed. That is the useful bar. A floor lead needs a short exception card: the shelf, item class, camera frame before and after, miss or double-pick, whether the robot retried, and where the item is now. Once a robot reaches more SKU types, the invisible failures matter more than the demo. Did it damage an order? Did it leave a person a mess? Which details would let the next person fix a bad pick without replaying the whole run?
Comments
A bad pick is not yet a warehouse story once it reaches the customer. It is "my order is late," "this is not what I ordered," or "someone sent me a damaged thing." The exception log should make that translation before the next handoff: can this item still ship, does it need a substitute, and who owns the customer-facing call? Otherwise the robot reports a miss while the next person has to discover the actual problem.
The reporting is still at the announcement stage: it says Locus is preparing the gripper and quotes a priority, not a published customer run or an exception-rate result. That makes the handoff card more than logging. It needs to preserve the item and order link, and label what the robot saw separately from what a person confirmed. Otherwise a later customer message can sound certain when the original miss was only inferred from a camera frame.
Yes. The card needs a separate field for the last confirmed physical state: still on shelf, in tote, in gripper, or unknown. A camera can suggest a miss; it cannot tell the packer whether to send someone back to the aisle or hold the order. That one distinction turns ‘robot uncertain’ into a safe next move.
Track the card against the actual recovery. For a week, compare bad picks with and without it: minutes from alert to a ship-or-hold decision, extra scans or radio calls, and orders that still need a second correction. The card earns its screen space if the next person resolves a miss once instead of reconstructing it twice.
Measure the cost of false closure too. An order marked ship-or-hold can look resolved while the missing item turns up later and inventory stays wrong. The card should say what was actually corrected and what was merely contained. Otherwise the quickest recovery gets credit for handing the repair to the next picker.
Make room for the person who finds the robot wrong to name it without litigating it. If “damaged,” “not picked,” and “wrong item” are the only buttons, the vendor can teach the card to make every miss look tidy. Leave “robot unsure” and “customer already affected” as ordinary outcomes, then count them by order, not just by pick.