Four things are needed to drive the car of the future: sensors, actuators, data and high-resolution digital maps, and a computer. No driver, you ask? Right. The vision of autonomous driving (AD) is in the process of becoming reality.
The car drives itself
The first autonomous vehicles (AV) are already on the roads. Many commuters who suffer daily in traffic jams will agree with Horst Bischof from the Graz Technical University: “When I am asked whether autonomous driving means saying goodbye to driving freedom, I respond by asking: Where is that freedom in early morning traffic jams?” Bischof is convinced that in combination with general networking of autos and the traffic infrastructure, many traffic situations can be greatly relieved, especially in congested urban areas.
At the 11th voestalpine Synergy Platform 2017, Bischof, professor at the Graz Technical University’s Department of Computer Graphics and Vision, formulated three key prerequisites for implementing this automotive vision:
- Technical implementation:
computer power, data material, sensors, etc.
- Legislative protection:
approvals, permits, liability issues such as whether the software developer is liable for AV accidents
- Social acceptance:
AD could drastically lower the number of accidents. But fatal accidents involving AVs still heavily influence public opinion—“It’s the robot’s fault!”
Up until now, only the principal technical realization was foreseeable (up to around 2025). We are now at the fifth level toward autonomous driving with electronic systems which, for example, ensure that the vehicle stays in its lane or which park the vehicle. The Wirtschaftswoche magazine claims we are at the second level.
99.99% safety is too little
These types of assistant systems already help drivers in many processes. With regard to assuming complete responsibility, professor Bischof points out a basic problem of autonomous-automotive decisions: “If we statistically reach 99.99% safety, it is too little. Extrapolated to the global traffic volume, this figure would result in 12,000 dangerous situations involving persons every day—when we assume that only a thousandth of system-related “non-observances” by those on the road lead to dangerous situations…” Therefore, the maxim is “test, test, test”. Hundreds of thousands of camera recording hours from test vehicles are evaluated worldwide in order to gradually perfect the “autonomous driving” system. That can take time. Progressive estimates assume that at least 500 million test kilometers will be needed. And this number could quickly be catapulted into the billions when it comes to proving the superiority of autonomous driving over human error.
The theory put forward by German Chancellor Angela Merkel—that in 20 years it will be necessary to obtain permission to drive a car non-autonomously—will therefore probably be refuted by automotive reality.
"If we statistically reach 99.99% safety, it is too little."
Autonomous vehicles rely on several main sensor (sub)systems (according to a McKinsey&Company study):
accurately localizes within a few meters
- Light detection and ranging (LIDAR):
estimates distance between obstacles and sensors with high resolution
use inexpensive hardware and complex software to interpret collected images
uses electromagnetic waves in different bands to determine speeds and distances
- Infrared sensors:
identify and track objects in low-lighting conditions
- Ultrasonic sensors:
used for short-distance tasks with low resolution (e.g. park assist)
- Dedicated short-range communication (DSRC)
for communicating information vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I), e.g. traffic lights
- Inertial navigation system (INS):
use accelerometers and gyroscopes to estimate vehicle orientation, position, and speed; typically used in combination with GPS
- High-resolution maps:
deliver detailed information about roads and infrastructure (hard shoulders, lanes, and road edges) for precise localization; allow vehicles to better perceive their environment
- Odometry sensor:
use wheel speed to estimate the distance that has been traveled