BMW collecting data from owners
Four in five drivers give BMW permission to use their driving data for future model developmentBMW developers have confirmed they are using a pool of more than 1.2 billion kilometres of real-world data from customer cars to design their next generations of software and hardware.
The project manager of the new BMW 7 Series limousine, Christoph Fagschlunger, has admitted BMW has been collecting data from their customers’ cars for the past three years – but only with their permission.
The data is being used to develop everything from the next generation of driver-assistance features to learning what is actually used most in infotainment systems, he confirmed.
The system involves participating BMW cars sending packets of data back to the German car-maker’s technical centre in Munich every time the cars are switched off, the engines are shut down during an idle-stop or, in the case of EVs and plug-in hybrids, the vehicles are charging.
Fagschlunger expects to have more than two billion kilometres of anonymous real-world data drawn down by the end of this year.
“We only do this with the permission from each customer – they have to give it – and nearly 80 per cent of customers do so,” the engineer explained.
“With this approach we have gained about 1.2 billion kilometers out of the customer fleet.
“It shows us things like where the active cruise is used, the lane control, how is it used, is there degradation of the systems and is the customer happy to use it?”
Some of the data, he admitted, has led to heavy conversations with product planners and designers about what technology is really necessary to develop compared to what customers actually use.
“The customers with Professional version of the navigation use active cruise control for 60 per cent of the whole drive time, which surprised us,” Fagschlunger said.
“But the assistance for the steering, lane keeping, is just 36 per cent. Some customers order the full set-up, but then, in the modes, they turn the steering assistance off.
“We are analysing why. It’s totally different on the geo-regions and we have to go through the data to determine if it is because of dense traffic, or road conditions, and then that is used for data-driven improvements.
“We see all this, and we also see how heavy is the usage on the sub-features.”
BMW stores the data in the car, then sends it over a simple but secure mobile connection from the vehicle to itself – paid for by BMW – to access the information.
“On average a normal car in normal use would send data three or four times a day, and the cost is covered by BMW
“It does not store much. It’s only kilobytes. The camera is optimised to 10KB or something very small. That’s how Mobileye created their good results with Roadbook, and they did it with BMW vehicles as well, and have used it to stitch together a 16-hour loop of virtual test driving with BMW vehicles,” he said.
One of the keys to the data is accessing it without having to trawl through millions of lines of information, and BMW uses crowd-data collector scripting to catch its ‘fish’.
“The developer can decide what is relevant to him or her and what they want to get answered out of the fleet data,” said Fagschlunger.
“So he can say, from tomorrow at 8:00am onwards, I want to find all the 5 Series vehicles with V8 engines in New York taking right turns at traffic lights and what do they see in that situation.
“They can then clearly have a filter on whatever they want, and not only big data.”
The real-world data has driven BMW to switch up its Level 2 and Level 3 driver assistance programs from acting like a computer driving on the road to behaving more like human drivers, without breaking the law.
“With driver assistance at L2 and L2+, it is so relevant to have a data-driven approach because you can’t do it all with logarithms and algorithms. It has to be optimised so that at the end it is more competitive with the owner for driving time because it feels like a normal driver, and not an intelligent system,” Fagschlunger said.
“That has to be the case. If you talk about a more urban use case, it’s not the case where, as we all once thought, that it’s best to have the best geo position on a high-definition map, but it’s more the question ‘are you driving with the car in the position a normal driver would use?’
“It’s more ‘Are you using the same trajectory on the junction?’ or ‘Are you handling the speed while accelerating?’
“That can be gained out of the customer fleet.”
Related: Massive data stream for cars of the future
Join the conversation at
Or email us at
Or email us at
Tags
Please see our Editorial Guidelines & Code of Ethics (including for more information about sponsored content and paid events). The information published on this website is of a general nature only and doesn’t consider your particular circumstances or needs.
