For those that have trouble with German lol - Upon reading this, I'd have to say it's probably not us @Rarosch unless I'm missing something? - "Autobrains, a start-up from Tel Aviv, is said to have developed a revolutionary different approach.As WirtschaftsWoche learned exclusively from investor circles,Continental, the first major supplier group, wants to useIsraeli technology in its systems for autonomous driving"
Anyway the article from @Rarosch is below for everyone to read:A new approach could give artificial intelligence for self-driving cars the decisive impetus.A German automotive supplier has already secured the revolutionary technology from Israel.But Facebook is also working on it.
After controversial debates, the German Bundestag has finally passed a law for autonomous driving.This creates the legal framework for cars that make their way through traffic on their own.Technically, however, the breakthrough has been a long time coming, despite huge industrial investments.It is true that individual providers, such as the
Googlesubsidiary Waymo or the company Cruise, which belongs to
GM, can now show impressive partial successes.Taxis with Waymo technology, for example, drive driverless passengers in some major American cities from A to B - but only in a geographically narrowly defined area.A universally applicable, secure system is still missing.But that is now within reach.
The key role in its development is played by artificial intelligence (AI), which derives and executes the correct driving maneuvers from the data from sensors such as cameras, radar, ultrasound or lidar.One problem so far has been the enormous effort that is required to train this AI.For example, image recognition algorithms, which are of central importance in autonomous driving, have to be trained with billions of images in countless repetitive loops until they clearly identify certain objects or living beings.
For example, pictures of children playing or stop signs are shown to the AI system until the system recognizes them 99.9999 percent reliably.In difficult borderline cases, such as pictures of crooked signs with graffiti disfigured or brightly clad, motionless children in front of a light background, people in low-wage countries like the Philippines now manually tag millions of pictures with individual semantic markers.The AI is trained further by repeatedly showing it such marked images: "That too is a child."
Autobrains, a start-up from Tel Aviv, is said to have developed a revolutionary different approach.As WirtschaftsWoche learned exclusively from investor circles,Continental, the first major supplier group, wants to useIsraeli technology in its systems for autonomous driving.On Wednesday,Conti presentedits plans to a group of investors atBank of Americain New York.The supplier had already extensively tested the Israelis' system and could largely confirm its advantages in practice, according to the investor conference.
EvenFacebookreports recent successes
The new approach has been researched for a long time and works under the name Unsupervised AI, in German, for example, “not guided AI”.In contrast to the conventional approach, the AI programs should themselves develop reliable criteria for object recognition and refine their algorithms.The advantages according to the Conti presentation: The development of the new AI systems gets by with a tenth of the previous amount of data and computing power.This would significantly reduce development times and costs.
Former Conti boss Karl Thomas Neumann sits on the Israelis' supervisory board.“Unsupervised AI is extremely exciting because it calls into question the entire mainstream of current AI developments relating to autonomous driving,” he says.If the concept catches on, self-driving cars will be more quickly deployable across the board because, for example, they can better cope with new situations for which they are not yet explicitly trained.
Autobrains, which, like many Israeli start-ups, was founded by former AI specialists from the local military, is not the only company researching unsupervised AI.A few days before the Conti presentation, Facebook's AI team surprisingly announced an important step in the new approach.TheFacebookresearcherswritein their in-house blogthat they have made “great progress in automated image recognition”."For the first time, Facebook's Unsupervised AI software has succeeded in independently identifying objects and people in videos without this software having previously been fed corresponding information about the objects from outside."
The system works in a similar way to the human brain when it comes to seeing, according to the Facebook researchers, in which the brain also only obtains a small part of the data required for object recognition from the optic nerve and generates the majority itself based on certain criteria.Experts speak of object segmentation here.So far, it has been considered one of the greatest unsolved tasks on the way to self-learning camera systems, as it could also revolutionize autonomous driving.
(20min delay)