Mobility of the future the driverless car is still stuck in a technology jam

Even though automated driving is making significant progress, driverless autonomous driving is still a long way off.

No scenario, no technology inspires the idea of the everyday world of the future like autonomous driving: self-driving cars, also known as robot cars, transport people from A to B in a relaxed and accident-free manner, find parking spaces independently and pick up passengers at the agreed time at the agreed place again.

As beautiful as this idea is, it is far removed from the future. Until driverless autonomous driving really becomes a reality, development in the five levels subdivided – from assisted driving (stage 1) to semi-automated (2), highly automated (3), fully automated (4) and autonomous driving of stage 5. The current focus is on bringing highly automated level 3 driving to the road. And for shuttle services and so-called robo-taxis, stage 4 is currently being developed.

Driver assistance systems pave the way for autonomous cars

Driver assistance systems play a central role in autonomous driving. Parking assistance, lane departure warning and adaptive cruise control, to name but a few, are already in widespread use today and are intended to improve the quality of life for drivers driving safety improve. For the manufacturers, the aspect of increasing comfort is probably not entirely unimportant. And the issue of better economy, i.e. lower energy consumption, will certainly gain in importance in view of the climate debate.

According to the U.S. market research firm expert market research, the global market for advanced driver assistance systems (ADAS) promises to grow at a rapid pace over the next five years average annual increase of 17 percent. This means that the volume will grow from around 25 billion dollars last year to 65 billion in 2026.

This is proof that the technology works well and is constantly improving. Example adaptive cruise control or adaptive cruise control (ACC): the system detects cars ahead, determines distance and speed and adjusts the driving behavior, i.e. distance and speed, of the own vehicle accordingly. Such systems already work well and reliably today, but mainly in predictable situations, for example on highways. This also applies, for example, to a lane-change assistant that supports drivers when overtaking.

News from Fraunhofer IKS directly by mail SAFE INTELLIGENCE newsletter

• Quarterly, unsubscribe at any time
• New articles, background information
• information on events& dates

Still a lot of research needed on artificial intelligence

The situation is quite different when artificial intelligence (AI) comes into play as another pillar of autonomous driving. Driverless cars must also be able to unforeseen situations move safely. The prerequisite is reliable detection of the surroundings, especially of other road users. AI algorithms are responsible for this. However, the AI’s machine vision (perception or computer vision) is not (yet) reliable enough to be considered for use in autonomous vehicles on public roads.

This is the area in which the Fraunhofer Institute for Cognitive Systems ICS is conducting research. Under the keyword safe AI the scientists are working, among other things, on making the uncertainties of AI-based perception quantifiable, for example in difficult traffic situations, in order to be able to meaningfully evaluate the behavior of perception. The fraunhofer IKS wants to demonstrably reliable systems by first assigning an interpretable value to the AI and making it visible to it. This is how the hitherto non-transparent classification of artificial intelligence can be mastered.

Driverless cars need reliable environment models

Another area of research at the fraunhofer IKS in the field of autonomous driving is the Safeguarding of environment models the respective driving situation. Driverless cars continuously monitor their surroundings using various sensor technologies such as radar, lidar and camera systems. Based on the data collected in this way, sensor fusion algorithms and AI create an environment model, which in turn serves as the basis for all driving decisions.

But the individual sensors, depending on the driving situation, have weak points. Weather conditions, incidents on the road and specific driving situations can have a significant influence on the driving decisions quality of environment detection have. It is therefore all the more important to ensure that the safe operation of the self-driving car is guaranteed – despite these weaknesses.

The researchers at the fraunhofer IKS are therefore working on systematic safety analysis methods. these are designed to provide a reliable indication of the completeness of the risks considered and to optimize the system design in terms of safety, performance, reliability and cost.