Cycle Time ≤5s, Identification Rate 99.9%: Migration Technology 3D Vision De-palletizing Project for Cardboard Boxes and Totes

When 23 types of cardboard boxes, two types of pallets, and mixed positive/reverse stacking patterns are presented simultaneously, can the vision system still stably identify and accurately locate them? We found the answer in a recently delivered intelligent de-palletizing project.

In industrial settings, de-palletizing is never a simple "moving task". Cardboard boxes with different colors, similar sizes but varying orientations, pallet incoming materials mixed with 0° and 180° placements, and interleaved odd/even layer stacking—any uncontrolled variable can cause the automated production line to "stall".

Recently, the 3D vision intelligent de-palletizing project jointly carried out by Beijing Migration Technology Co., Ltd. and an intelligent technology company successfully passed acceptance. In the alternating operation scenario of two de-palletizing stations, the system stably guides the robot to complete high-precision grasping and placement of cardboard boxes and totes, with a camera cycle time of ≤5s per capture and an identification rate of ≥99.9%.

This project needs to handle two types of materials simultaneously: - Cardboard boxes: 23 specifications, minimum length/width dimension difference of only 15mm, stacking patterns include regular grid stacks and mixed odd/even single/double layers, and incoming material orientations include 0° and 180°. - Totes: Standard totes, stacking pattern is five-flower stack, also with mixed positive/reverse placement.

The camera needs to be installed on a moving mechanism, covering two de-palletizing stations, alternating between C and D stations for imaging, guiding the robot to complete grasping and conveyor belt unloading.

The core difficulty lies in: many types of workpieces, non-fixed stacking patterns, orientation to be determined, and grasping sequence has specific requirements—the vision system must not only "see" but also "calculate accurately".

The project uses Migration Technology's self-developed Epic Eye Log L 3D industrial camera, fixed on a moving mechanism, switching to positions above the two de-palletizing stations for imaging according to the control system's commands.

The camera is installed directly above the center of the stack area, ensuring complete coverage of the entire material stack. The vision system communicates with the robot via TCP, calculating the pose information of one workpiece per capture, and operates interleavedly with the robot's action cycle without affecting overall efficiency.

The entire system achieved high-performance output during on-site acceptance. For cardboard box materials, the identification accuracy reached ±5mm; for tote materials, the identification accuracy was ±3mm. The identification rate for both types of materials reached ≥99.9%, and the vision cycle time was ≤5s/capture, fully meeting the requirements of high-speed production line continuous operation.

During the debugging phase, the system demonstrated stable identification capability—even when the length/width dimension difference of cardboard boxes is only 15mm and the stacking pattern is grid stacks with consistent orientation, the vision can still stably distinguish; when incoming materials have 0°/180° mixed placement, the vision clearly outputs the orientation, guiding the robot to grasp in a fixed direction.

In the field of automated de-palletizing, there is a huge engineering gap between "being able to run" and "being able to produce stably". The reference value of this project lies in: - Extremely flexible, compatible with multiple specifications: One system compatible with 23 types of cardboard boxes + 1 type of tote, supporting multiple stacking patterns such as five-flower stacks, grid stacks, and mixed odd/even layers, without requiring significant program changes for changeovers; - Precise positioning, stable grasping: For material offset, varying box orientations, and slight surface defects, 3D vision adaptively corrects poses, combined with fixture error-proof logic, significantly reducing jamming, dropping, and collision; - Dual-station reuse, cost reduction and efficiency improvement: A single 3D vision system covers two de-palletizing stations, saving hardware investment and improving space utilization, suitable for compact workshop layouts; - Standardized delivery: Clear hardware list, standardized installation specifications, quantified acceptance criteria, and replicability.

Migration Technology has always focused on the implementation of 3D vision guidance technology in industrial scenarios. If you are planning intelligent de-palletizing, random picking, positioning and assembly, etc., welcome to discuss specific requirements with us.