A proof of concept at KLM Cargo at Schiphol shows that a robot system developed by DERO Groep, guided by AI vision software from Fizyr, is capable of processing virtually all packages from aircraft pallets, despite wide variation in packaging types, sizes, and materials.
The air cargo sector is under increasing pressure. Growing volumes, tight labour markets, and stricter standards around physical workload are pushing air cargo operators worldwide to explore automation of warehouse and ground handling processes. Depalletizing — unloading aircraft pallets down to warehouse format — is one of the most labour-intensive and physically demanding steps in the chain. At the same time, it is one of the hardest to automate, due to the unstructured and highly variable nature of air freight: no two shipments are alike. While e-commerce and retail logistics have long benefited from automated palletizing with standardized packaging, the air cargo sector has lagged behind in this regard. The expectation within the industry is that AI-driven image recognition holds the technological key to closing that gap — provided systems are robust enough to handle that variability.
From aircraft pallet to europallet: heavy and repetitive
At KLM’s cargo hub at Schiphol, incoming air freight shipments follow a fixed process: first, aircraft pallets (PMCs) are unloaded, then packages are restacked onto europallets, after which a forklift handles transport to the next location in the warehouse. The middle step — depalletizing — is currently done entirely by hand. It is physically demanding work, particularly for packages exceeding fifteen to twenty kilograms.
“We want to support our employees by reducing physical strain,” says Yu-Ming Sie, Innovation Lead at KLM Cargo. That principle formed the motivation for a proof of concept with an automated depalletizing system, developed by DERO Groep in collaboration with Delft-based AI software company Fizyr.
Wide variation as a technical challenge
What makes the application at KLM Cargo technically demanding is the extreme diversity of incoming freight. Dick Goudriaan, Business Development Manager at DERO Groep, describes it this way: “The pallets coming out of the aircraft have an incredible variety — in dimensions, packaging materials, strapping, you name it. It is very unstructured, which requires a great deal of understanding through the camera and translating that into an action for the robot system.”
The robot system DERO developed for this application consists of three robots — two for depalletizing and one for palletizing — and a multifunctional gripper capable of gripping both from the top and from the side. That flexibility is not a luxury: during the project it became clear that packages fall into two main categories. Cardboard boxes can be picked from above, but packages with a plastic outer layer require a grip from below. The gripper must be able to perform both operations reliably.
AI vision as the link between camera and robot
The robots are guided by Fizyr’s perception software. “Our software allows us to understand what we see through a camera and translate that into a trigger for a robot or an automated process,” explains Tibor van Melsem Kocsis, CCO of Fizyr. The AI determines not only what is on the pallet, but also the exact position and orientation of each package — information that is essential for the robot to execute the correct gripping movement.
Results exceeded expectations
The outcomes of the proof of concept are encouraging. Goudriaan: “We found that 97 percent of the cargo can be picked automatically.” That figure was higher than KLM Cargo had anticipated going in. “We did not know what would come out of it, because the solution did not yet exist,” says Sie. “We learned that the robot could process a high percentage of packages — much more than we expected.”
For KLM Cargo, the test answers the central question that underpinned the project. “The key question was: can the technology help us today? And the answer, as we have learned, is yes — the technology is there.”
