Many key technologies, such as autonomous driving, are on the verge of a breakthrough. Practical problems still stand in the way. With the "Learning Chips Lab" research project, three professors at Fachhochschule Dortmund - University of Applied Sciences and Arts have found a way to solve all these problems. They are currently presenting this approach at the Hannover Messe.
When an autonomous driving car encounters a motorcycle, the car must do several things:
- Recognize that it is a motorcycle.
- Know what that means in traffic.
- Assess whether a reaction is necessary.
- If so: decide which is the best response,
- and react.
And all this permanently, within fractions of a second, and even if there are other cars, cyclists and people on foot in the vicinity. This is possible thanks to the latest developments in artificial intelligence.
One of the most important aspects is absolute reliability. So if the Internet connection were to weaken and the calculations got stuck in the cloud while the car and all the other vehicles continued to whiz through the area, that would be bad.
Independence from the cloud
The dependence on the cloud to date is a crucial weakness in autonomous driving just as it is in many areas of Industry 4.0, energy technology, biomedicine and mobility as a whole. In addition to the unreliable connection, data protection and the enormous hunger for performance are also major problems with cloud computing.
That's where the idea of the Learning Chips Lab comes in. Using the possibilities of machine learning, information technologist Prof. Dr. Hendrik Wöhrle is developing methods that enable computers to plow through huge data sets on their own, for example, to understand what a motorcycle is and reliably identify it in any traffic situation.
These chips are absolute specialists
Together with electrical engineer Prof. Dr. Michael Karagounis, who also designs chips for CERN's particle accelerator, among other things, he is designing computer chips that are optimized for such applications - and that are so powerful and at the same time energy-efficient that they do not rely on cloud connections when in use, thus simply bypassing the aforementioned problems.
Finally, computer scientist Prof. Dr. Carsten Wolff of the IDiAL UAS Institute adds a special twist to these developments: it must be possible to program the new types of chips in such a way that users do not have to familiarize themselves with the chips' complex technology.
And all this open source
Contrary to the industry's habit of keeping all details of their achievements secret in order to avoid imitations, the researchers at Fachhochschule Dortmund have opted for the greatest possible openness and are publishing all results of the research project as open source, supervised by Prof. Wolff. In this way, scientists all over the world can take up the results and develop them further, thus stimulating and accelerating this research immensely.
The "Learning Chips Lab" started in the fall of 2021 and is funded as a research focus by the Ministry of Culture and Science NRW.