Artificial intelligence –
not so difficult after all

The blue boxes are positioned. A small camera records the contents – all ready for the test. A tense Reda Jaber and Nicolas Hummel turn their heads towards the monitor. A green check appears: All in place – the boxes are filled exactly as they should be. Smiles on their faces. What seems unspectacular at first glance is technologically highly sophisticated: In Volkswagen’s Smart.Production:Lab in Wolfsburg, IT specialists Jaber and Hummel are watching artificial intelligence (AI) at work. The experiment shows: AI can independently check whether delivery boxes are packed correctly – in the laboratory or in genuine logistics scenarios.

Industrial Computer Vision is the name of the system that Reda Jaber and Nicolas Hummel have developed together with a seven-member team and in collaboration with other labs and Volkswagen’s Software Development Center in Dresden. The basic idea: people train artificial intelligence to evaluate optical data and detect errors – extremely reliably and in fractions of a second. The AI can check boxes for completeness, for example – but many other applications are conceivable. “The user interface is so simple that anyone can operate it and train the AI independently. You don’t have to be a computer scientist,” says Nicolas Hummel. Over the next few years, Volkswagen expects Computer Vision applications to generate savings in the double-digit million-euro range. The focus is on production and logistics.

Industrial Computer Vision offers a toolbox for the implementation of AI use cases: All you need is a person who prepares enough training material – for example, photos with correctly and incorrectly packed boxes – and marks them accordingly. “The AI then learns independently to distinguish faultless from faulty results. After just a few hundred training images, you can achieve good results,” says Hummel. A few days are often enough to make a neural network ready for use. Artificial intelligence – not so difficult after all with Computer Vision.

The experts at the Wolfsburg Smart.Production:Lab do not usually develop promising new applications on their own, but in cooperation with other competence centers in the large Volkswagen Group. Take Computer Vision, for example: while the software is written in Wolfsburg, the Data Lab in Munich is concerned with improving neural networks. The Software Development Center in Dresden tests the system in practice and provides feedback. With success: In a logistics hall in Wolfsburg the automatic container control system is already in use. People in the images are automatically pixelated to comply with data protection regulations – a solution that will be standard in all Computer Vision applications.

“Our preferred solution is an online offering from which each department can download the appropriate applications.”

Reda Jaber

Another possible application is currently being tested at Porsche in Leipzig: Here, AI checks labels that are attached to every car during production. The labels contain vehicle information or instructions for airbags, for example, and are written in various languages. AI checks the images in real time and provides feedback as to whether everything is in order. From next year, the application is to be used at other locations – rolled out via the Industrial Cloud, which links Volkswagen Group’s plants together.

In the Smart.Production:Lab, Reda Jaber and Nicolas Hummel are also continuing to work on making the advantages of Computer Vision available throughout the entire company. “Our preferred solution is an online offering from which each department can download the appropriate applications as if they were in a library,” says Reda Jaber. A pilot test is already underway.

And when everything is ready? Jaber and Hummel know they won’t run out of ideas. In addition to Computer Vision, the experts at the Smart.Production:Lab have also been working on driverless transport systems, data analysis, robotics and the Internet of Things (IoT) – all major topics for the future. 

“It’s fun to be able to help shape all this. I learn an incredible amount because we develop almost everything ourselves.”

Nicolas Hummel

Reda Jaber takes a similar view of the work in the Smart.Production:Lab: “What I like best is the freedom to try out new technology.” Flat hierarchies allow the individual teams to organize themselves independently and also have a say in selecting the most appropriate projects. “Anyone can contribute ideas, all employees are heard,” says Reda Jaber. This was also the case with the work on Computer Vision – the success proves the developers are right.