Where does AI deliver tangible value in the warehouse? We speak with CIM CEO and AI expert Hannelore Mayr about her experience and insights from research and practice, as well as the potential AI offers for modern warehouse and WMS solutions.
Ms. Mayr, artificial intelligence is currently a major topic in logistics. Where do you already see measurable value from AI in intralogistics today - and which use cases do you consider more of a vision for the future than an operational reality?
In an industrial context, AI can currently be used wherever it is “safe enough”. Large language models (LLMs) can serve as explanation assistants that, for example, explain configurations or analyses.
Image recognition can simplify data collection. We have a customer that uses an app for item recognition. Based on a photo, the app provides us with the item quantities, which are then prefilled at goods receiving.
Optimization algorithms are gradually moving in this direction as well. These include machine learning (ML) and reinforcement learning (RL) algorithms that can, for example, enable predictive maintenance for conveyor systems.
AI-based methods also make data analysis comparatively straightforward. This makes it possible to identify potential for optimization. LLMs help reduce complexity.
There are certainly other use cases in robotics and automation technology. New types of robots are now available whose functionality is optimized using AI. We recently integrated a pallet unloading robot from Copal.
PROLAG World now includes a digital twin for automated warehouse environments. What potential do you see in combining digital twins and AI—for example, for simulations, bottleneck analysis, or optimizing material flow strategies?
To train a machine learning (ML) or reinforcement learning (RL) algorithm, you first need a large amount of data. Naturally, real-world data is the first choice because it ensures that the data reflects reality. However, if you want to train algorithms for situations that do not occur frequently, real-world data reaches its limits.
One example is our anomaly detection system in the automated warehouse. The ML algorithm predicts whether a pallet is likely to emerge more slowly than others or whether problems are likely to occur during processing at the K-station, potentially causing pallets to circulate.
In reality, anomalies do not occur that frequently. To train the algorithm properly, we reproduced various situations in our simulation environment and generated sufficient data of different types.
Another major advantage of the digital twin is its testing environment. We need to test the algorithms extensively. This is not possible exclusively on the customer's system. Parameters may also need to be adjusted. Using the simulation environment, we can determine whether the predictions work.
CIM has been involved in university research projects for many years. What needs to happen for a promising research result to become a stable, reliable, and economically viable function within standard WMS software such as PROLAG World?
A very important aspect is our customers. We try to work closely with them as early as the research projects themselves. For example, customers provide us with test data, or we talk to customers who have relevant use cases related to the topic in question.
Once we move beyond the initial phase, the focus shifts to becoming more concrete. To this end, we have already conducted test sessions for AI applications with customers.
Of course, investment is also a factor. We have been investing heavily in our standard software for years. The AI field - including ML, RL, and the integration of LLMs - is also an important area in which we invest so that our customers can become even more efficient through our software and digitalization becomes easier for everyone.
A WMS currently plans, controls, and monitors warehouse processes. Will AI lead such systems to increasingly identify optimization potential independently, prepare decisions, or dynamically adjust processes in the future? And where do you see the limits of autonomous warehouse control?
The WMS as such will always exist as long as there is a physical need for it. What we must not forget is that AI in intralogistics is the result of data being aggregated beforehand. This data is aggregated through warehouse management—in other words, collected by WMS systems.
Even if all WMS logic could eventually be replaced by AI, the WMS would still ultimately do the same thing: control inbound and outbound movements and provide an overview. It would simply do so even better and in a way that is more tailored to the customer than with conventional logic.
In our research projects, we have seen that current AI logic can be used very effectively to complement existing algorithms. Training a machine learning algorithm to ensure that hazardous goods cannot be stored in certain storage locations is far more complex than programming this rule.
By contrast, conventional programming is much more complex when it comes to making different decisions based on data, because there are often many special cases. ML and RL algorithms can handle these situations much better.
AI is already helping us today, and it will continue to help us manage the volume of data and complexity even more effectively in the future. I believe that ML and RL algorithms, as well as LLMs, will help us digitalize logistics faster, make fewer errors, and deliver higher quality in the future.
One example is identifying optimization potential. We have already implemented a monitoring system in the automated warehouse that reports delays at an early stage. A similar approach can also be envisioned for manual operations.
We also see potential for AI in optimizing processing - for example, when calculating picking lists based on historical data. ML and RL algorithms can provide support, particularly in complex decision-making situations that are difficult to fully represent using fixed rules alone.
Our research projects have shown for many years that new technologies are particularly useful when they deliver tangible benefits for warehouse logistics. Many insights from research therefore flow step by step into our warehouse management software. This allows us to test new approaches at an early stage under realistic conditions and develop solutions that genuinely support our customers in their day-to-day warehouse operations.
