Speaking to the Australian motoring media at the Frankfurt motor show, Klaus Fröhlich, member of the board of management at BMW, said that his company now uses more sensors and more hardware for Level 2 autonomous technologies than some competitors (such as Tesla) have for Level 5.
“It’s not possible to put sensors in place for Level 5, because they are not available yet. We have to develop chip-based laser scanners, [and] we don’t need just one - we need five,” Fröhlich said.
According to Fröhlich, the ability to mimic the human eye and brain to accurately judge a situation while driving requires enormous computing power that is not yet available or easily tested.
“The most fickle thing, is the environment model. What you see with your eyes, it’s really perfect, we can see what is living, what is dead. We can predict pedestrians, but in a car when you’re driving down a street with 50 pedestrians, we have to calculate the most probable way they would go for all the pedestrians or bicycles... its challenging, so for the moment the tech is not available to do fully autonomous driving from our approach to be really safe, because a level 5 accident means we are liable.”
The main problem for developing such a technology is the required testing, which Fröhlich says is significantly more than what any company has had to do before.
“On a 5 Series I needed less than five million kilometres to develop [the new] 5 series. On autonomous driving Level 5 I have to do 150 million test kilometres and I can’t do it on the road, so we have to build a very big computing main frames, up to 500 petabytes to calculate 95 per cent of our driving in virtual [environments].
BMW is also in a fortunate position whereby its cars have been connected with SIM cards for a number of years, meaning it is actively collecting data with the sensors already placed in existing cars, helping increase its data pool to better develop autonomous driving technologies.
Even so, the challenges faced with today’s mobile networks are not sufficient to bring about the speed and data requirements needed for fully autonomous vehicles.
“At the moment we have LTE, but we need 5G in 2020. Fully-autonomous driving where you have to download very detailed maps, requires 5G.”
According to Fröhlich, it’s also vital to have a fast network in case of a hacking attack.
“If you have a car it’s very important that we have no hacking attacks, so we have an approach that our cars are always connected to our backend... [then] if there is a hacking attack, I can do an encryption on 10 million cars within a day.”
That doesn’t mean, however, that BMW’s autonomous cars will rely on the network. In fact, they are designed to work in any condition.
“The car has to be able to work without a network, the car loads the high-definition map for whatever regions of 20 kilometres and then it can continue to drive.
"A fully-autonomous-driving car, in our concept, has 33 sensors, from laser scanners to long radars up to a lot of cameras because cameras are very good to define human beings or animals… So the car can drive in the first failsafe mode without being connected.
“Perhaps you reduce speeds, but the car has to always work. Even if a sensor breaks down, you can’t… if you’re driving on a German motorway at 250km/h, you can’t brake hard, it’s not a good idea, so we always have a step-by-step approach to reduce functionality so the car can always, in the end, park somewhere.”
BMW is building, at the very minimum, at least two levels of redundancy and sometimes up to four levels for each of its components.
The company expects fully-autonomous Level 5 driving technologies to begin mass-market adaptation in the 2020s. However, even for now, it is building such vehicles as prototypes with a philosophy that it would start at Level 5 and then scale back to lower levels so that its architecture is correct from day one.