Pushing the Limits of #SelfDrivingAI with Uwe FrankeJune 12, 2017
It’s long been a dream of engineers and drivers alike to one day sit in the back of a car, relax, and allow the vehicle to drive itself.
From a young age, this has also been Daimler’s Head of Image Understanding, Uwe Franke’s dream too.
Since 1989, Franke has explored and developed intelligent vehicle innovations for Daimler, including lane departure systems, image understanding, and stereo vision – the latter of which was the foundation for Daimler’s Stereo Camera-based safety systems found in mid- and upper-class Mercedes-Benz models today.
Franke shared his vision for the future of intelligent driving and image recognition this week at Mercedes-Benz Research & Development, North America’s (MBRDNA) “Machine Learning and Self-Driving Cars” meet-up at the Sunnyvale facility – bringing hundreds of engineers and partners together to discuss the future of self-driving cars.
“In 2013, Mercedes-Benz brought the S-Class to market – the first production car with automated driving features that actually looked like a car,” Franke remarked, explaining how early prototypes looked like “computers on wheels.”
Through the development of Daimler’s revolutionary Pedestrian Recognition capabilities, Franke and team gained a deeper understanding of the world, recognizing different scenes in traffic such as nighttime, rainy conditions, and distinguishing pedestrians from inanimate objects. Daimler’s image understanding capabilities track amongst others motorcycles, bicycles, rods, poles, and traffic signs.
Today, pixel-level performance has increased over time from approximately 65.3% to over 80% – in fact, Daimler’s real time label system can identify more than the control human labeling, and with more accurately.
Franke acknowledged hesitations and concerns about “human” issues such as limited vision at nighttime and small objects in the road, so Franke and his team tested their early detection reliability systems in-person.
During the test drives the system sometimes detect obstacles that even the driver did not see. Both in daylight and at night, the system was able to identify those obstacles better than the engineers could.
Franke ended his presentation with the question on everybody’s mind – given this leading technology are we ready for autonomous driving.
Indeed, the appetite for autonomous driving cars is growing, with customers asking more and more for cars with computer assistance. The saying goes, “no road is long with good company,” and whether we are ready for such intelligent company in the driver’s seat remains to be seen.
Adding to Franke’s foundation of mapping technologies, Wendy Ju, Executive Director at Stanford University Center for Design Research, delved into how we interact with our cars on a human level, and the inherent limitations engineers face.
At Stanford, Ju’s team uses simulators to design and research how people interact with vehicles of the future – specifically, how they explore the emotional experience of automated driving, and the opportunities for learning and adapting to them.
“Surprisingly, distraction becomes engagement when it comes to autonomous driving,” Ju remarked. “If people are just ‘supervising the car,’ they will get bored and potentially fall asleep at the wheel.”
Ju also found it only takes a minute or two for drivers to feel safe enough to give up full control of the car – which means engineers need to find something for them to do so they stay alert while the computer drives the car.
While creative solutions are currently being explored, including implementing questions for drivers to respond to while the car drives to allow the computer to “learn” more about them, Ju’s team will soon be testing and replicating her studies in other places around the world. Everyone in Silicon Valley, she notes, is familiar with autonomous driving, so reaching other people who may not know as much about the topic will be important.
Both Franke and Ju have set up an incredibly important foundation for machine learning and understanding human behavior in the regards to self-driving cars, and both convey their optimism for technological advancements in the near future in making autonomous driving a reality.
Make sure to follow @MBRDNA and be the first to know when the next meet up will take place! Interested in learning more about MBRDNA’s research and development work? Stay tuned to this blog and our social media channels at Facebook, Twitter, LinkedIn and Instagram.