Automated Driving Cars Could Learn Anything

It’s just a question of resources. Development engineers at ZF are working to keep tight control of the spiraling volumes of data needed to master this challenge.

The ZF test vehicle glides smoothly along. Traffic is thin on the highway near Düsseldorf this morning. The test vehicle is equipped with three driver assist systems: an adaptive cruise control system, a lane-keeping assist system and a lane change assistant. “I’m now activating the system and entering the speed I want,” says Dr. Hans-Gerd Krekels, Vice President Active Driver Assistance Systems Engineering at ZF TRW, and takes his hands off the steering wheel.

The future has begun. We have taken a leap forward. Already useful and usable, active cruise control, lane-keeping assist and lane-change assist are all forerunners of automated driving, the megatrend defining today’s automotive industry.

Many of the innovations currently being developed to achieve this goal are already contributing to road-user safety and will eventually become standard equipment in the high-volume segment, just like airbags and intelligent braking systems in the past. ZF has already incorporated electronic control of some of the new safety functions into a Safety Domain ECU (electronic control unit). The ECU is installed in the test vehicle that is now cruising down the highway with Krekels at the wheel.

Portfolio for Automated Driving

Dr. Karl-Heinz Glander and his team are working on the software needed to realize the ultimate goal of automated driving

Dr. Karl-Heinz Glander, Krekel’s colleague at the Düsseldorf facility, is Senior Engineering Manager Automated Driving & Integral Cognitive Safety. At the facility, a team of some 30 developers is working on the software needed to realize the ultimate goal of automated driving. Why is ZF plowing resources into making cars that are able to see, think, and act? Glander explains: “Like the auto manufacturers, ZF is embracing the trend toward automated driving. As suppliers, our top priority is to ensure that our components are installed in those future vehicles.” And Krekels adds: “We have the entire portfolio of technologies for making automated driving a reality.”

The ECU is already helping drivers by controlling multiple driver assist systems. “If your car is fitted with emergency braking assist, you may already have experienced a situation where, without it, you would have plowed into the car in front at an intersection,” says Glander. The lane keeping system and blind-spot warning system also enhance road safety. Similarly, adaptive cruise control cuts fuel consumption, and automatic parking provides added convenience.

The ECU acts as the central integration platform, processing millions of bits of data generated by the environmental sensors, and analyzing the vehicle status and surrounding traffic. Links with the steering, braking and driveline systems make it possible to combine and precisely coordinate a whole raft of vehicle functions.

Although Dr. Hans-Gerd Krekels is not touching the steering wheel, he is alert and ready to intervene in an emergency as currently required by German law.

Auxiliary control unit

“The Safety Domain ECU (SDE) is an auxiliary control unit that makes it possible to integrate any number of disparate software packages. It’s a control unit for automated driving that operates within a relatively complex network of ECUs. It is not, however, a central control unit as such.” As things stand, there is no central control unit. Instead, there is a distributed network of control units made up of sensors, auxiliary control units like the SDE, and actuators. Brake control units and airbag control units are already fitted with powerful ECUs. All these individual units have sufficient computing power to perform specific actions.

Dr. Hans-Gerd Krekels, Vice President Active Driver Assistance Systems Engineering

All ECUs share certain features, such as multicore processors that are able to run multiple processes in parallel, and the ability to interact with third-party software through an open-standards interface. But there is no uniform strategy for assigning specific tasks to, for example, the mechanical actuators, or for deciding what should be done centrally and what should be done locally by the individual control units.

The processors used for this were originally developed for computer games. Since then, they have found their way into the automotive world. “The chips that can do this are currently non-safety chips from the world of infotainment – a world that was already tackling the problem of delivering sufficient performance to cope with graphics over 10 years ago,” says Krekels. As driver assistance systems become more sophisticated, the volume of data that has to be processed in vehicles will snowball. Competing concepts for dealing with this problem exist around the world, and a 100% standardized solution is unlikely to emerge. “The underlying technology – deep learning – and associated neural networks capable of mimicking the relevant human experience do not necessarily have to be centralized. It depends on how much computing power we need. We’re already starting to integrate this distributed approach into sensor systems.”

From ADAS to autonomous driving

Krekels and Glander like to compare the conventional Advanced Driver Assistance Systems (ADAS) installed in today’s cars with eyes that tell a hand what to do directly, without the intervention of a brain. These systems can only carry out one highly specialized task relating to a very specific case – such as staying in one lane, or not running into the vehicle in front. Glander explains: “Automated driving, on the other hand, is more akin to the way humans react. It aggregates collected information to describe the surrounding world. Existing ADAS systems don’t do this. This ability to describe, understand, perceive and interpret the world – that’s what’s new.” This is the difference between conventional driver assistance systems and emerging premium ADAS functions such as automated driving. “Automated driving is already cognitive, but it still isn’t as good as human drivers. We want to enhance computers so they’re as good as humans.” asserts Krekels.

Hence the intensive testing of driver assistance systems and interconnected ECUs in the test vehicle, explains Krekels as he exits the highway. His ultimate vision: a standard of automated driving that will significantly reduce the risk of accidents and remove driving stress in the long term.

Video: Testing ZF technology for automated driving

Adaptive cruise control, lane-keeping assist and lane-change assist: watch the video to find out how ZF’s intelligent sensor systems support automated driving functions.

Pictures: Carsten Behler, Mareike Foecking

Further related articles