Products for Passenger Cars
Artificial Intelligence Artificial Intelligence

Artificial Intelligence: The Algorithm as a Chaffeur

Min Reading Time
Artificial Intelligence (AI) is gaining ground in everyday life. This also applies to autonomous driving in road traffic. ZF innovations support this development.
Andreas Neemann,
author_image
Andreas Neemann wrote his first ZF text in 2001 about 6HP transmissions. Since then, the automotive writer has filled many publications for internal and external readers, showcasing his passion for the Group's more complex subjects.
Since the industrial revolution, there has been a clear division of tasks between man and machine. The latter score with their strength, precision and their enormous workload. Humans distinguished themselves through their intelligence and ability to think logically. This cooperation between man and machine is now changing: Artificial Intelligence is the keyword. Machines and computers are increasingly equipped with cognitive skills so that they can solve corresponding tasks just as well and often better than humans.

Deep Learning Makes the Difference

Deep Learning Makes the Difference

Artificial Intelligence has long since found its way into our everyday lives. Translation programs are used billions of times a day. They are also based on AI algorithms such as spam filters or plagiarism checkers. Many professions use big data analysis of intelligent software: Doctors use it when analyzing an MRI and X-ray images, lawyers when searching for precedent judgments and tax advisors when organizing client profiles.
Unlike conventional software, these AI systems today can make decisions or answer questions that are not already thematically stored in their code. This abstraction is made possible by Deep Learning. Algorithms simulate the synapse nodes of the human brain. Input passes through several layers, one after the other. The number of involved nodes is immense: Current AI systems organize billions of artificial neurons in roughly 30 layers. Only in recent years have processor performance, internet bandwidth and data available via the cloud reached the necessary level to support such self-learning programs.

The Algorithm Learns How to Drive

The Algorithm Learns How to Drive

Deep Learning algorithms allow feedback and correction loops. This quality is crucial when it comes to applications such as autonomous driving. After all, there are an infinite number of situations in road traffic that cannot be programmed in advance – especially when human error is also involved. Even the comparatively simple entry into a roundabout can be complicated: Another car, which is already in the roundabout and has set the turn signal, simply drives on. Experienced drivers instinctively recognize the dicey situation and avoid the accident.
The key difference from the traditional software world is that AI systems can make decisions and answer questions that are not already thematically stored in their code.
Arnold Schlegel, development engineer at ZF and expert in autonomous driving and AI

What about AI? Just like humans, AI can draw conclusions from circumstantial evidence: The speed of the other vehicle, the position of the wheels, the driver's line of vision – they all indicate a deviation from the thousands of vehicles that actually drove out of the roundabout when they set the turn signal. Conclusion: This one car will not turn despite the turn signal being on. While a conventional software would turn into the roundabout and cause a collision, AI will brake. Once the algorithms have been "trained" well enough, they detect such hazards even more precisely and reliably than humans, and react more quickly. After all, depending on concentration and mood, even the most experienced driver can be distracted from time to time.

An important aspect for autonomous driving is therefore the training of AI systems. This is also accompanied by the major challenge of validation: How can a system designed to solve problems that arise unexpectedly be tested? Virtual training and software-in-the-loop methods will make a contribution to overcoming this hurdle soon.

Artificial Intelligence Needs Computing Power

Artificial Intelligence Needs Computing Power

In order to be able to hand over control of the entire vehicle to AI in the future, the electronics architecture in the car must also fit. All systems must be brought together in a central control unit. This requires sufficient computing power to be able to evaluate the overwhelming amount of data from cameras, LIDAR, radar and other sensors in real time. This is made possible by high-performance control units such as the ZF ProAI developed by ZF. Its latest generation handles up to 1,000 trillion operations per second (TOPS), meeting the strict safety requirements for automotive applications. This latest version of ZF's automotive supercomputer will enter volume production in 2024.
The Group is thus setting an important course for autonomous driving. AI should soon help to significantly reduce the number of accidents. Once more and more autonomous vehicles are on the road, it will be possible to network them into an intelligent traffic management system. Then AI could even prevent traffic jams.

Milestones in AI Research