Driverless cars? Years away, says Mercedes

Mercedes-Benz’s chief engineer says the auto and tech industries underestimated the difficulty making driverless cars viable and safe
- shares

Auto and tech companies which made bold assertions driverless cars are right around the corner underestimated the difficulties in producing them, and the industry should now be understanding how complex and remarkable human brains really are behind the wheel.

That’s the position of Mercedes-Benz’s recently appointed board member for R&D, Markus Schäfer, who acknowledged this week that so-called ‘level 4’ and ‘level 5’ driverless vehicles are some fair way off the ubiquity many brands suggested was imminent, even on highways where fewer variables exist.

“We all probably underestimated the challenges in front of us, absolutely. It’s really all about the last percentile in safety, all about that. Performance has to be much better, it has to prove to be better than a human driver,” Schäfer told us this week.

He said despite Daimler AG’s six years of work across two continents and billions of euros invested, the hurdles remained huge.

The list of challenges goes on, and this senior-most engineer ought to know.

We’ve been told again and again by the likes of Google’s Waymo, Elon Musk, and Nissan that driverless cars (presumably level 4 and 5) are imminent. But Schäfer’s more measured words suggest the world’s top luxury brand is taking a more measured approach.

Mercedes’s program includes Level 3 highway partial autonomy on highways in the 2021 S-Class pending homologation and approval, an improbable (and on-track) joint-venture with arch-rival BMW to make potentially level 4 (highway) cars by around 2024, and a longer-term project with Bosch to produce geofenced robo-taxis... eventually.

Here is an edited (but contextual) transcript of our extremely interesting discussion with him in Frankfurt this week:

“Sensor tech is key for autonomous, especially level 4 and 5. It’s basically all about sensor fusion, and lots of AI software. LiDAR is key. The question is how far can you see and what can you see? It’s our philosophy to offer quality solutions and safe driving behaviour... so sensors have to be reliable and to look very far, and to look up steep angles,"

“... The example of one US company that you might have in mind... the bubble on the roof is a LiDAR, but one that costs as much as a complete car today. So as well as improving performance, visibility, range, angle, precision, it’s also about cost. Everything; hardware, software and system learning, has to be improved and that’s part of the development of level 4 and 5. This will be something that’s going to move.

“There’s always new technologies coming to take it to the next level, it’s a moving field. And right now there are some limitations on how far you can see with your sensor equipment, which is linked to the maximum speed you can drive. Imagine an obstacle in your way on the interstate, worst case a person, you need obviously to stop before hitting it.

“... It’s ready for production and you can produce a system but the system is somewhat limited... we would wish performance in a normal urban environment or highway environment would be better, it cannot (yet) cope with all situations that normally would occur... these limitations is why level 4 or 5 is moving a bit slower than anticipated.

“It’s very hard to predict, obviously the challenges are way bigger than anybody thought... from my point of view it’s going to take a couple more years, then you might have some usage in some restricted areas, maybe on a highway which is an easier case and will definitely come earlier.

"If it’s an urban environment it would be limited routes. Free-floating routes in all areas, that’s not very near.

“Is 99 per cent certainty enough? 99.9? That makes a huge difference. That’s company philosophy. We would err on the safe side, definitely.

“... We all probably underestimated the challenges in front of us, absolutely. It’s really all about the last percentile in safety, all about that, absolutely, performance has to be much better, it has to prove to be better than a human driver.

“For my team its a challenge to sort, probably underestimated by the auto industry and the entire tech world... now we understand what a human being is capable to do and to imitate that is a huge undertaking.”

Your thoughts? This writer thinks it highlights the philosophical divide between auto engineers grounded in mundanities like basic safety, and the futurist types who prefer bold pronouncements and stretch targets.