ZF presented the latest generation of its ProAI central computer for cars at the auto show in China – which, by the way, was held as a real trade fair with human visitors instead of a virtual event. These computers take over the control in the automation of driving – in a very wide range from level 2 (automatic parking, traffic jam assistant, etc) to level 5 (robot taxi without a human driver). Since these autonomy levels place very different tasks on the computers and their performance, ZF emphasises the scalability of their design. Most of these computing platforms will probably be based on corresponding pre-configured systems from Nvidia, but ZF emphasises that depending on the performance requirements, SoCs from other manufacturers will also be used. The new ProAI generation will go into series production in 2024 – coincidentally, this coincides with the availability of Nvidia’s next-gen HPC platform Altan, which was unveiled last week.
The performance data – up to 1000 TOPS – is also in line with what Nvidia is touting. The power consumption is 3 TOPS per watt, which, however, then requires active cooling, in the case of the most powerful models even by liquid. However, according to ZF, this is in any case significantly lower than the electrical power consumed by the current generation – according to ZF, power consumption has been reduced by 70%.
The ProAI is designed to fuse the signals from all the data-heavy all-round sensors of future generations of cars and process AI algorithms to make decisions.
With all these features, ZF’s ProAI is strikingly similar to its counterpart from Continental, the current generation of which is already installed in Volkswagen’s ID.3 and ID.4 electric cars.
In addition to the ProAI, ZF showed a system for automated, driverless parking (valet parking) in Shanghai that, in contrast to a system presented earlier by Bosch, works independently of the infrastructure. ZF’s vSLAM (Visual Simultaneous Localisation and Mapping) relies on a high-resolution front camera, a front radar, four surrounding cameras and twelve ultrasonic sensors. Additional sensors can be configured as well as the connection to a backend.
Competitor Continental – which, as is known, has a conceptually very similar HPC in its range – did not focus so much on the computer hardware at the show in Shanghai, but already demonstrated the next evolutionary step: The interlinking of the on-board computer with the cloud (see also separate article). At the Shanghai show, Continental presented an end-to-end connectivity solution vehicles. This functional integration covers the entire data path from the sensor via the domain or central computer and the networking unit to the cloud.
Continental: Cloud is king
The low-latency connection of the vehicle network to the outside world shown enables real-time processing of all data entering the vehicle and processed by HPC computers in the vehicle. Software functions are distributed between the cloud, the high-performance computer and so-called zone controllers. This approach enables a high degree of flexibility and functional safety at the same time. To a limited extent, functions from driver assistance can also be integrated into the system. Continental’s self-developed 5G mobile radio module (Network Access Device) offers scalable computing power with improved cyber security.
The module also supports various hardware options to meet the requirements of different customers and regions. The system allows flexible system integration options for car manufacturers and suppliers – from stand-alone network access devices to full integration into telematics ECUs or smart antenna modules. The supplier is already working with several car manufacturers for 5G telematics based on their 5G Hybrid V2X platform.
Continental also showed how far development has progressed in the meantime in sensor technology development, especially radar. At the Shanghai event, the company showed a new, more powerful generation of surround radar and long-range radar sensors with ranges of 200 and 250 metres respectively and high resolution. With their compact design and high sampling rates, the interconnection of several such sensors enables 360-degree all-round coverage. The new radar sensor solutions are thus intended to enable predictive applications for Euro NCAP requirements up to higher automated driving functions, such as automated lane changes.