The AI ​​gets smarter with every trip

The scene takes place in the heart of China, where KURIER and other Austrian media take a test drive in a driverless taxi. The Metropole Wuhanwhich had gained questionable fame five years ago as the hotbed of the corona pandemic, became the first city in the world in 2021 in which the so-called robot taxis have left the test phase and are traveling completely without a human monitor. Today, 400 of them already shape the cityscape.

The taxis are operated by Apolloa subsidiary of the Chinese search engine giant Baidu, which also has the necessary KI-Software developed. To do this, the group uses cars from Chinese suppliers – in this case a car from the brand Arcfox – and equips them with the in-house software. Users in China can book rides via the Apollo app and then unlock the car using a code.

So while the skyscrapers on the banks of the Yangtze River pass by outside, the steering wheel turns by itself. Otherwise, the driving experience is unspectacular: it’s reminiscent of a driving instructor showing a new student what it looks like when you meticulously follow all the rules. The AI ​​always stays at least a car length away from the car in front. She brakes early but leisurely, taking every curve without momentum. And it stays within the speed limit even when overtaking.

AI researcher Liu Hao from AI Research Center in Wuhan explains how this works: “AI models can’t really make decisions independently. They just predict the potential danger of every possible action and recognize this more quickly than humans – then they always choose the action with the least danger.”

Only when the Apollo taxi detects many danger spots at once, for example on streets where pedestrians and cyclists are also traveling, does it steer suddenly to increase the distance. This is particularly noticeable when another car swerves out again Robo-Taxi illegally overtaken from the right. It is an open secret that many Chinese drivers are “testing” robo-taxis by intentionally putting them in dangerous situations to potentially provoke a rear-end collision.

A risky tactic – after all, Apollo’s cars are equipped with at least twelve cameras and record every journey. If another road user is at fault, Apollo calls it a “passive accident” and reports it. Only if the collision was triggered by a decision by the AI ​​is it considered an “active accident”. According to Apollo, there has never been one like this before.

Despite daring maneuvers by other drivers, everything went smoothly during the test drive in Wuhan. For 28 minutes and 12 kilometers you only need to pay 2.50 euros. Human competition cannot keep up.

“We believe our work will take humanity as far forward as the moon landing,” says Apollo headquarters. “Our technology will fundamentally change the way we get around.” So how close are we to fully automated city transport?
“It would already be technically possible for all cars to be self-driving,” says Liu Hao. “But the technology will be implemented gradually so that the population can get used to it.” The Apollo program will gradually be expanded to other major cities.

Self-driving taxis also drive along the wide streets on the outskirts of the capital Beijing, in the modern industrial district of Yizhuang. But they are not alone here: AI has also been experimenting with for years in a huge glass office complex that would put many a spaceship to shame. It is the headquarters of Jingdong, or JD for short, the largest online retailer in China.

Its website is a kind of Chinese Amazon, but JD also offers many in-house products – and the promise of same-day delivery almost anywhere in China if the order is received before 11 a.m.

High tech everywhere you look

An employee explains how this is possible during a tour of the high-tech exhibition hall on the ground floor. Although JD has an army of more than 300,000 suppliers under contract, most importantly, almost all of its warehouses are fully automated.

This means that an in-house AI takes over the logistics, self-driving transport vehicles and robot arms handle the goods. These vehicles can also find their way to their destination independently and react to the busy traffic in the warehouses. People are only employed there to monitor the machines.

The group has thus reduced its costs and increased efficiency. There are already fully automated JD warehouses in Europe: in Germany, France, Poland and the Netherlands. However, the Chinese do not want to deliver from there themselves (yet). Instead, they rent the locations to competitors such as Amazon or Ebay.

That should change in the future. JD soon wants to enter the market abroad under the name “Ochama”. It is hoped that the company’s internal research department will continue to achieve success: In parts of China, JD is already using around 700 driverless transport vehicles for deliveries, as well as large, propeller-driven drones that can fly up to a ton of goods. The delivery route has already covered around a million kilometers.

Smarter with every trip

That’s what’s special about artificial intelligence: it gets smarter with every new trip. “The intelligence of an AI increases with the size of the data set on which it is based,” explains Liu Hao. “At the moment we can constantly add more data, just like our western competitors. That’s why we haven’t reached a plateau in AI development yet.”

Europe can only dream of all this. While the road traffic revolution advances on Chinese and even a few US roads, the EU states are being left behind. And at least the Apollo AI is already ten million kilometers ahead of us.