Start with the incoming pallet, not the camera catalogue. If the pallet position, carton dimensions and layer pattern stay within a known pick window, a mechanically located pallet and a fixed sequence may be the simpler first trial. When cartons arrive shifted, mixed or in changing layers, the cell may need to locate each available top carton before committing to a pick. The choice turns on the variation actually delivered to the station, plus whether the gripper can hold those cartons reliably.
Separate a stable pattern from a stable pallet
“Same carton every day” does not mean “same pick position every day.” Compare several incoming pallets rather than a neatly built demonstration stack. Record the pallet’s placement at the infeed, carton dimensions, layer offset, overhang, damaged corners and the order in which top cartons become reachable. A fixed sequence is credible only if the total variation stays inside the gripper’s usable contact region and the robot’s clear approach path. The permitted variation has to be defined from samples and the proposed end effector; there is no useful universal centimetre threshold.
If pallet placement itself drifts, mechanical guides may solve more of the problem than image processing. If the pallet is located well but the top cartons move independently, the question shifts to box localization. If mixed stock or broken layers are routine, test how a sensing option identifies the next available carton and how it recovers from an obscured or ungrippable one. Public depalletizing specifications distinguish defined single-item layers, varied defined layers and mixed pallets. These categories show why input patterns matter; choose from the buyer’s actual incoming flow.
The visible pick is a starting point for the trial
In the filmed demonstration, a white robot with a multi-cup tool lifts one carton and moves it between two pallet areas. That makes the top-contact and transfer arrangement easy to see. Use the buyer’s incoming pallets to test how much variation the proposed route must handle.
Whatever position strategy is chosen, carton grip is its own test. Cardboard is described as a porous, abrasive material for vacuum handling; the cup needs a workable seal against the actual surface. Use the customer’s real filled and empty cartons, including different tape patterns, labels, worn tops and box weights. Watch the pick, travel and release rather than counting only successful lifts from the middle of a fresh stack. A carton that can be lifted once may still slip during a turn or fail to release cleanly at the receiving position.
A decision trial for both routes
First run the stable-pattern candidate with the actual pallet locator and a sample of incoming stacks. Record each carton position at first contact and the number of times the tool needs a recovery action. If the measured spread stays inside the tested contact and clearance window, retain the fixed approach and simplify the specification. If it does not, identify whether mechanical location, carton presentation or sensing addresses the variation. Then test a sensing candidate against the same difficult pallets and explicitly define what happens when the next carton cannot be located or gripped.
This comparison avoids buying perception features to compensate for a pallet that can simply be located, and avoids promising fixed coordinates for a flow that rarely arrives in the same place. Neither route removes the need for a representative box-grip trial.
Send EVST sample cartons from different batches, photos or measurements of the incoming pallet patterns and the observed position variation. We can discuss a fixed-location trial and, if the variation warrants it, the conditions a sensing trial would need to pass. That gives the integration team a decision about the incoming flow before it selects a more complicated cell.