Putting AI behind the wheel

Dr. Ulf Wilhelm develops software for driver assistance and automated driving systems at Robert Bosch GmbH. He works to make driving safer in his role as Lead Architect Advanced Driver Assistance Systems and believes the trend toward increased automation in vehicles will affect far more than just how we drive.¹

For over 175 years, the people at ZEISS have asked the question: How can we challenge the limits of imagination? In celebration of that vision, ZEISS has partnered with thought leaders and great minds from around the globe for the ZEISS Beyond Talks, giving them center stage to speak about their own work, visions, passion and issues that are affecting our world moving forward.

My background is actually in physics, more precisely quantum electrodynamics. You might wonder how I transitioned from physics to the automotive industry. In physics, the focus is on building predictive models and verifying those predictions against real-world phenomena. This is reflected in my current work, which involves creating models of human drivers and anticipating how they will act in the real world.

One of my primary motivations is the dream of consistently reducing accidents. Mobility is a fundamental freedom that allows us to travel wherever we want, but it comes at a significant cost. I believe that we must work to reduce that cost.

Passive safety measures, such as airbags, play an important role, but they are not sufficient on their own. Real change will come from intelligent vehicles – specifically, those powered by artificial intelligence (AI). My hope is that we can achieve a significant reduction in automotive accidents through these advancements.

At the core of a driver assistance system is an artificial driver. It is designed to understand what actions should be taken in various driving scenarios and involves a lot of different sensors, such as radar and cameras, which perform measurements of the environment. The system senses, understands and acts based on its surroundings.

A camera generates a vast array of pixel data in a two-dimensional number field. The first real challenge is to identify the objects within it – to go from 1,000,000 integer values to, "There is a pedestrian on the right who wants to cross the street." This is easy for people, even a three-year-old can do it. But developing algorithms that can do the same thing requires extensive data processing and advanced AI.

Once we identify the actors in a driving scenario, the challenge is to predict their behavior over the next few seconds. This is particularly complicated in an open-world environment – there are countless scenarios, and we cannot enumerate all possible reactions. Additionally, computational power limits how many possibilities can be analyzed in real-time.

On top of this, this analysis must occur 20 to 100 times per second to ensure safe driving.

The algorithms themselves are highly complex. We're talking about millions, even billions of parameters that are defining the learned algorithm. Neural networks help by spotting patterns that you do not see yourself. Using large amounts of data, machine learning (ML) finds patterns in behavior so the algorithms can react to really complex cues and predict the behavior of other vehicles or pedestrians.

We are moving towards integrating numerous control units within vehicles. Currently, a typical vehicle may contain 50 to 60 control units and around 150 million lines of code. The target is to consolidate these units into only a few, powerful vehicle computers. This will enable the implementation of complex algorithms and AI technologies in decision-making processes.

I believe that what we are doing with driver assistance and automated systems is only the beginning. A car is just a robot acting in an open world. If we manage to achieve Level 5 autonomy, the implications extend beyond mobility. We could develop machines that perform tasks currently only possible for humans – fully automated and unsupervised.

The journey toward advanced driver assistance systems and full automation is not just about improving transportation. It is about leveraging AI to create a safer, more efficient and ultimately more connected world. The advancements we make today will lay the groundwork for a future where machines can enhance our lives in ways we are only beginning to imagine.