Announcing Nicole Beringer as our Keynote Speaker – GoZinc 2018

Automated driving needs a novel approach to cope with driving scenarios that are currently solved only by the driver’s control. The reasons for the driver to take over control vary from limitations of the range of ego sensors or recognition algorithms to required information, e.g. infrastructure information like traffic lights, that cannot be derived from in-vehicle sensor observations. What all have in common is that any reaction, from the driver as well as from a driver assistance feature, needs to be on time. This becomes clear when looking at the range of ego sensors (e.g. LiDAR sensors about 40m ahead). The driver may want to have the speed reduced in advance before a speed sign is reached or be warned in time to take over control if the autonomous driving road ends. It is crystal clear that it needs more than just a high-quality in-vehicle sensor processing in order to obtain a wide range of HD information needed for automated driving.

To make vehicles react properly in these kinds of situations the classic embedded software development needs support by self-learning algorithms.

Our approach combines techniques well established in robotics like Simultaneous Localization And Mapping (SLAM) as well as end-to-end protection and image compression algorithms with big data technology used in a connected car context. This allows extending the sensor range as well as the sensor availability of a single vehicle. For behavior perception, another well-established technique in automatic recognition scenarios enters the game: deep neural networks (DNNs). Although DNNs rely on training data in detection situations, they perform much faster as traditional software. The more samples and the more iterations the DNN gets, the better accuracy is obtained in all classification and detection situations.

DNNs also become a good alternative to validate functional safety as well as the safety of the intended functionality.

The keynote speech will analyze the different approaches and scenarios of using SLAM and DNN algorithms in automated driving and will give an impression how embedded software development for automated driving benefits from robotics and neural networks.