I have long been quite sceptical about the use of AI in warehouse operations. Almost all tasks that can be performed using AI, think problem solving and prediction, pattern recognition or decision making, can also be performed automatically (and with guaranteed high quality) with classical tools. Think statistics (pattern recognition and prediction) and operations research models (optimal decision making and prediction). Or they can be performed manually and efficiently, using the right IT.
However, I must confess that I have gradually had to change my opinion. There are more and more applications where AI plays a major role that work very well. Those applications can be found in a growing number of vendors' products and are used in an increasing number of operations. I give five examples below.
Product identification and classification
When a new product is received, it needs to be registered in the master data file before it can be processed. To do this, it has to be measured, weighed, the supplier has to be looked up, and so on. This is time-consuming work, despite the use of all kinds of tools. With the help of 3D cameras and AI-based pattern recognition, all these things can be entered into the system at once with great accuracy. This leads to time savings, which is especially useful in companies where the product range is constantly changing.
Stock counts
It was only a few years ago that Walmart removed Bossa Nova robots from its shops, due to insufficient quality of counting and price recording. But the successful applications are there now. Meanwhile, drones can do this work automated and accurately in pallet warehouses. 3D cameras register pallets and can automatically count the number of boxes and thus the stock, with excellent precision. In warehouses with broken pallets and packages, the technology does not yet work accurately enough, but that too may change in a few years.
Robotised order picking
About a decade ago, the first pick robots appeared on the market. These robots could pick individual products from a bin. However, the performance was not impressive. Many picking errors (product could not be picked up properly, or fell out of the gripper), it was slow, and basically an operator had to be constantly on standby to fix errors. This has changed. The speed of identification and gripper analysis (relative product position, texture and centre of gravity analysis) has gone up, thanks to 3D cameras and fast AI. Robot speed has also improved. The percentage of incorrect picks has been greatly reduced. Combined with parts-to-picker systems, picking robots can therefore now be used increasingly cost-efficiently.
Safety on the shop floor
Last year's Logistica Award winner Essensium detects risks on the shop floor with 3D cameras and a communication system between warehouse trucks. Collision risks are analysed with AI and truck speed is automatically adjusted. Conversely, truck speed can also be increased in case of safe situations, leading to higher efficiency. This application is just one example; there are now many systems that can monitor processes and improve safety.
Stacking colourful pallets
Stacking furry pallets is still a challenge. The goal is to stack (robotised) as many variegated products as possible stably on a pallet (or roll container), taking into account sequence and other stacking constraints. The number of combinations is large and, in addition, each box can also be rotated, making the number of possibilities even greater. AI plays a role in identifying the box and deciding its orientation. Although mainly OR algorithms are currently still being used, I expect a lot from AI-based pattern recognition that can potentially make faster decisions about stacking.
These are just five examples. There are many more applications and they are getting better and better. Even sound OR algorithms can increasingly be replaced by AI-based decisions based on pattern recognition. The quality of these decisions is increasing. Knowledge of OR is no longer necessary. Programming a good machine-learning algorithm is also becoming increasingly easy (e.g. using AI, such as ChatGPT, or Deepseek).
Of course, AI also has drawbacks. The best-known disadvantage is that a lot of, varied, and unbiased training data is needed for the AI algorithm to make good decisions. If the decision is outside the domain of the training data, the AI decision will be poor. However, good data is increasingly available as AI learns from new implementations.
In short, I'm bowled over. AI is also unstoppable in warehouses. It will lead to better processes, more efficiency and a safer working environment. However, despite all the robotisation and AI, we will continue to need people, at least for now, to do the manual work in warehouses.
