Startup improves AI training procedures for autonomous cars

Startup improves AI training procedures for autonomous cars

Technology News |
By Christoph Hammerschmidt

algorithms must first be trained. This learning process requires a very large number of images. During this learning process, the images are marked, a process known as labeling. “An algorithm learns from examples, and the more examples there are, the more effectively it learns,” explains Philip Kessler, one of the two founders of For the training, the car industry needs a lot of image and video material for machine learning for autonomous driving. The objects on the pictures are traditionally marked by hand. “Large companies employ thousands of workers in low-wage countries; the process is laborious and time-consuming,” explains Kessler. “At, we use Artificial Intelligence, which makes it ten times faster and more accurate to do this marking.”

Although the image processing process is largely highly automated, the final quality control requires the use of human labor. Kessler emphasizes that the combination of technology and human care is particularly important in safety-critical topics such as autonomous driving. The markings in the image and video representations, also known as annotations, must match the real environment with pixel accuracy. The better the quality of the processed image data, the better the algorithm that trains with it.

“Since it is not possible to provide training images for all situations – accidents, for example – we also create simulations from real data,” says Kessler.

For the initial phase, the company wants to focus on the topic of autonomous driving. In the future, also plans to process image data that will be used to train algorithms for tumor recognition or the evaluation of aerial photos. Leading automobile manufacturers and suppliers in Germany and the USA are among the customers of the start-up founded by Kessler together with Marc Mengler in 2017. In addition to its headquarters in Karlsruhe, the young company has more than 50 employees in Berlin and San Francisco. In 2018, a round of private investors provided it with start-up financing amounting to 2.8 million US dollars.

More information:

Related articles:

New neural network training approach cuts energy use, time

Baidu launches ‘no code’ AI model training platform

Simulation enables training of deep learning algorithms in virtual driving tests

Nvidia puts a lot of effort into autonomous driving


If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :    eeNews on Google News


Linked Articles