Düsseldorf So far, there has been a bottleneck in the development of autonomous driving that is slowing down the development: Training Artificial Intelligence (AI). Currently, this is mostly done manually. People all over the world evaluate pictures of recorded test drives. They use them to mark all objects by hand so that an AI can recognize them at some point – an enormous effort in terms of personnel and time. Continental now wants to accelerate this process with its own supercomputer.
In cooperation with Nvidia, the supplier has developed a computer that can be learned faster with large amounts of data based on AI. According to Conti, he is also able to create simulations that can partially replace test drives. The supplier’s supercomputer is one of the 500 most powerful computers in the world and is located in Frankfurt.
With this, Conti is taking further steps to further reduce the manual portion in the development of autonomous driving. The supplier had already taken part in the Israeli start-up Cartica AI a year ago, which uses AI to automatically classify certain objects on the test drives recorded – for example, cars or pedestrians – without human intervention.
Continental currently drives around 15,000 kilometers a day with its fleet – and that generates around 100 terabytes of data. One of the tasks of the Nvidia computer is to evaluate this mountain of data. The neural networks of the AI systems are also trained with the information. They mimic the way the human brain works and are intended to ensure that a vehicle masters a lane change or a roundabout like a human being.
“The supercomputer is an investment in our future,” says Christian Schumacher, Head of Program Management Systems in the Driver Assistance Systems business unit at Continental. “The state-of-the-art system reduces the time it takes to train neural networks because at least 14 times more experiments can be carried out at the same time.”
Tesla is also working on its own supercomputer
With its own supercomputer, Conti even has something ahead of Tesla. The US electric car manufacturer hides his own plans for a supercomputer behind the project name “Dojo”, which is supposed to evaluate images for autonomous driving with the help of neural networks. Tesla boss Elon Musk Dojo mentioned for the first time in August last year. Tesla has apparently not yet used such a supercomputer.
Conti also takes a different approach than Tesla. Because, according to Musk, simulations that the supplier would like to use would not be suitable for training a robot car AI. The reality is too complex to simulate a computer, Musk said in a video message at the World Artificial Intelligence Conference in China in early July.
Among other things, this should also be due to the amount of data available. Because Tesla’s autopilot is already in use and is busy collecting video material that is sent to the headquarters in Fremont. Almost five billion kilometers have already been covered. The data come from around 50 countries.
Only the Google subsidiary Waymo should have similar amounts of data. Suppliers such as Bosch or Conti and also the German car manufacturers are lagging behind here. It therefore makes sense for Conti to use traffic simulations to make up this gap.
For Nvidia, working with Conti is the next step in increasing its influence in the auto industry. “The project was set up on an ambitious schedule and was implemented in less than a year,” says Schumacher. “After intensive testing and the search for suitable companies, we chose Nvidia, which equips many of the fastest supercomputers in the world.”
In addition to Conti, almost all suppliers in the field of autonomous driving cooperate with the graphics card manufacturer. At times, Tesla had also used Nvidia’s chips. At the end of June, Nvidia also announced a close cooperation with Daimler. From 2024, the central computers of all new models of the car manufacturer are to be equipped with chips from Nvidia. The Mercedes operating system should then run on the computers. They also control all connectivity and autonomy functions.
More: These are the five hurdles on the way to autonomous driving.