autoevolution
 

Mercedes-Benz Wants Its Cars to See like Humans Do

We've mentioned this before, even a couple of days ago: the future of driving will change irreversibly, and the biggest alterations to the act of controlling a car will happen during our lifetime.
Image labelling 1 photo
Photo: Daimler AG
The biggest disruptive innovation in the automotive sector will definitely spawn from the growth of active safety. If systems like ABS and ESP made the first big step toward the creation of active safety technologies, the third millennium is about to completely modify how we perceive their help (or intrusion, depending on who you ask). The exponential advancements in computing power are taking their toll on R&D expenses, though, which is why only the richest carmakers are currently on track to bring the biggest innovations.

Thankfully, Daimler AG doesn't have a money problem, and some of the most disruptive innovations in terms of active safety may very well come from the Stuttgart car giant. In fact, Mercedes-Benz is already testing a couple of pre-production technologies that will transform the car from a luxury means of transportation into a personal butler and bodyguard on wheels, all in one.

As many of you city folks have noticed, driving in major urban areas has become increasingly difficult in recent years, since you need to pay attention to pedestrians crossing the street and checking their smartphones, cyclists that think they own the road, unsupervised children and city traffic in general.

Sure, some modern cars already have auto-braking features, with more expensive ones even managing to distinguish between a person and dog, for example. Mercedes-Benz wants to take those systems up a notch, and it recently achieved a fantastic breakthrough in connection with the UR:BAN research initiative, which is backed by a German consortium of other carmakers, universities and companies.

Scene labelling

Using a camera-based system that technically “sees” in 3D and that can simulate human binocular vision, the technology developed by German engineers can automatically detect all types of objects and even classify completely unknown situations. Researchers from the “Environment Sensing” department fed tens of thousands of images from various German cities to the system. The photos themselves depicted the most common city driving situations, while the researchers had manually labelled 25 different object classes, including vehicles, cyclists, pedestrians, buildings, lampposts or trees.

The system then automatically learned to detect any object that can obstruct driving in any way, even if they are partly hidden, while also adapting the car's auto-braking features depending on a situation that hasn't yet happened. The computing power available to the system is what makes this all possible. Essentially, a bunch of deep learning computers were neurally networked in a similar manner as the human brain, so the system learns from past mistakes and acts as an undeveloped artificial intelligence.

The testers

Daimler researchers have so far built five different test vehicles, each of them showcasing different parts of the near-production technology. The first car demonstrated the aforementioned scene labelling, managing to distinguish between all sorts of objects in various urban environments, making the car act accordingly and not hit anything that crosses or may cross its path. In other words, this is autonomous driving taken to the next level.

The second test prototype showed how radar and camera technology can work together to visualize and resolve both dynamic and static objects, essentially being capable of continuously building a 3D map of all the surroundings near the car while it's driving in an unknown area. Both size and speed movement attributes of every object can be scanned and introduced into an environment model, which also allows incomplete sensor data and missing information and can also work in fog and bad weather.

The third vehicle was fitted with a system that can detect, classify and identify not only pedestrians and cyclists, but also guess their next move. Just like a (good) human driver, the system takes into account body position, head posture and position on the road or sidewalk and can then predict whether a pedestrian will cross the road in front of the car or stay on the curb. According to Daimler researchers, this would make autonomous braking systems trigger up to one second earlier than today.

A fourth test car showed how the stereoscopic camera and the radar-based system can make autonomous lane-changing in heavy city traffic, assisting the driver in a speed range between 30 km/h (19 mph) and 60 km/h (37 mph). Using situational analysis, the system can predict how a driving scenario may develop and, therefore, can allow or not the car to make the intended lane change. It should be noted that the driver can intervene via the steering wheel, accelerator or brake pedal at any given moment.

The fifth test vehicles demonstrated how it can sense and predict driver behavior before a planned lane or direction change. Most of us first check out the mirrors, glance over the shoulder and usually slow down before initiating a turn. The system takes all those details into account and cross-references them with the map information and steering movements. Therefore, if it senses that you're about to make a turn, it can activate the correct turning signal automatically to inform other drivers and road users of your intention as early as possible.

Instead of a conclusion

Times are really changing, and the work in a production-version of the aforementioned system is almost done. Considering it's based on sensors that already exist in a lot of current Mercedes-Benz models, we wouldn't be too surprised if something similar could debut as early as next year, in the 2017 E-Class W213. If not, the Stuttgart carmaker has already pledged that it will have fully autonomous models on the road by 2020, and those will definitely use most of the breakthroughs working in the all-seeing test vehicles referenced in this very article. We should all get ready for the future because it's right around the corner, and it has a fluffy baseball bat in hand.
If you liked the article, please follow us:  Google News icon Google News Youtube Instagram X (Twitter)
About the author: Alex Oagana
Alex Oagana profile photo

Alex handled his first real steering wheel at the age of five (on a field) and started practicing "Scandinavian Flicks" at 14 (on non-public gravel roads). Following his time at the University of Journalism, he landed his first real job at the local franchise of Top Gear magazine a few years before Mircea (Panait). Not long after, Alex entered the New Media realm with the autoevolution.com project.
Full profile

 

Would you like AUTOEVOLUTION to send you notifications?

You will only receive our top stories