Autonomous driving on the way to a mass market product

Autonomous driving - driving at SAE levels 4 or 5 - is progressing only at slow pace in Europe. This may have something to do with the strategic priorities of vehicle manufacturers. Nevertheless, it was also a much-discussed topic for the AEK 23. The biggest challenge is the crucial step from driver assistance systems to truly autonomous driving on a large-series scale, explained Mobileye CTO Shai Shalev-Shwartz. Whereas with ADAS systems - including the L3 systems already available on the market - it is essentially the human who drives and is supported by the machine, autonomous driving involves a role reversal: The machine controls the vehicle, while the human monitors it.

In order to successfully take this step and ensure maximum functional safety, Mobileye relies on an incremental approach from a methodological point of view and a modular approach from a technical point of view. This involves securing all decision-making processes through redundancy. In doing so, the Shalev-Shwartz design team relies on design principles that have proven themselves in the aviation industry for many years. A high degree of redundancy is achieved through multiple independent, different ways of acquiring and processing data. The systems must also be validated independently of each other. The goal is to achieve a "superhuman" level of fault tolerance: According to statistics, there is one fatal accident for every 3.75 million hours of driving on the road. The industry is aiming for a much higher error safety of 10 million driving hours. "However, this is a matter of legal regulations," Shalev-Shwartz said. The Mobileye CTO also provided some insights into the practical implementation of such a platform. According to his presentation, the Israelis' autonomous driving platform uses one main camera, several subcameras and a lidar sensor. There was no mention of radar.

Mobileye's solutions are installed in very many driver assistance systems, with on-site data processing always taking place with the help of the "Mobileye chip" within the camera. Against this background, the thesis "Centralization is not the right way" expressed by Shai Shalev-Shwartz in Ludwigsburg gains even more significance, which he did not even talk about in his presentation.

Robert Chu, head of the "Apollo" program at Chinese technology company and automaker Baidu, took a very different look at autonomous driving. As part of this program, Baidu has gained extensive experience with autonomous cars. Baidu vehicles are already being used commercially in complex metropolitan traffic situations. However, the company not only builds robotaxis as such, but also offers its technology to competing OEMs as a whole or in components. The offering includes the electronics platform as well as high-resolution map material for the Chinese road network. This map material is collected via crowdsourcing and continuously updated by the participants. Apollo's self-driving platform relies primarily on cameras and radar for its sensor technology; lidar sensors are also on board for redundancy reasons. AI mechanisms for signal processing and decision-making also play an important role. With the deployment of this platform at several Chinese OEMs, Chu expects autonomous driving to make a rapid breakthrough in China. "We are making this technology affordable for the masses," Chu said.