Auto Sapiens – How to design cars That see, think and act safely
Cars are about to make an evolutionary leap, developing artificial intelligence.
Vehicles still have much to learn from the human brain and its ability to handle complexity efficiently. The organ handles basic functions like breathing and walking more or less automatically, instead focusing processing power on situational awareness, decision-making and the coordination of more complex tasks.
Until recently, having three ECUs communicating with each other in a vehicle was considered complex. And while that may still be sufficient for certain functionalities in the future, it is no substitute for a human driver.
For automotive developers, replicating the human brain’s talent for managing complexity efficiently is now a major goal. Future vehicles need on-board intelligence that can deliver automated driving functionalities for consumers.
For cars to think and judge where to go and how to accomplish that – when to accelerate, brake and steer – a series of connected ECUs is not enough. Vehicles need an electronic architecture that creates an abstraction layer.
In short, vehicles need operating systems to sit between the hardware and the software functions. Download a smartphone app and the phone’s operating system takes care of the hardware dependency. The software developer does not need to install the sensors, graphics engine and processor.
Accelatering progress in this direction
Implementing this in a vehicle is more complex, but the automotive industry is accelerating progress in this direction.
Karl-Heinz Glander, senior engineering manager for Automated Driving & Integral Cognitive Safety at ZF, says: “Currently, if the driver wants an automatic emergency braking function, for example, the automaker buys the radar hardware along with the control algorithms. Updates happen when a new generation of hardware launches, along with new code. Automated driving will not work like this.”
"Cars that offer automated functions will have electronic architectures that will be independent of the hardware. With the arrival of Microsoft and Intel, development of specialized computers like the Spectrum ZX and Commodore 64 ceased. Vehicles are starting to make the same type of transition.”
In the past, a vehicle might have dedicated sensors for blind spot monitoring and for a Traffic Jam Assist function but no exchange of information between the two. With component-oriented architectures, even when components are on the same CAN network, combining these functions can be difficult.
“With more advanced electronic architectures, safety systems can evolve from making simple decisions based on just two or three signals,” says Glander. “When planning a future vehicle platform with 20 sensors, more and more automakers now look first at how to combine all that data into a single, centralized architecture. That could be with a central, powerful ECU or by sharing the computing power of different ECUs.”
Taking a functional approach will make it simpler for automakers to use environmental sensors to help improve crash protection. For example, if an approaching vehicle hits a patch of ice and starts to spin, an automated vehicle could predict the path of the oncoming vehicle.
That way the car could calculate an evasive maneuver and prepare the airbag and restraint systems for the potential collision.
“ZF’s advanced R&D is developing concepts for ‘integral cognitive safety’,” says Glander. “That means giving vehicles artificial intelligence, instead of just putting signals together. And it means defining vehicle functionalities – the properties and capabilities you want the car to have – before deciding anything else.”
This is a fundamental shift in the way the auto industry sees safety and the way it is organized. Today, the teams responsible for the hardware tend to develop systems in silos, backed by purchasing organizations that source hardware with software embedded.
The focus is now shifting to equipping vehicles with full systems, not components, with new capabilities. In the process, software is increasingly recognized as a product in its own right. These changes in the industry’s mindset will accelerate the development of artificial intelligence.