The 5 major challenges of ADAS and autonomous driving
What automotive players have to struggle with when it comes to develop and integrate the data logging and computing architecture to enable fully autonomous vehicles
In the journey to level 5 autonomous driving – the highest achievement in terms of driving automation – automotive players are encountering many challenges when it comes to develop and integrate the computing architecture to enable fully autonomous vehicles, as long as advanced driver-assistance system architectures (ADAS). Here are the main ones.
The quantity of data generated by an autonomous vehicle, even at moderate levels of automation, is unprecedented and unmatched by traditional embedded computers. Overall, the total bandwidth used by the sensors amounts to 3 to 4Gbit/s in the simplest case and 40Gbit/s in the most sophisticated ones, and it is expected that these data rates will increase during this phase of technological development. The nature of the application requires constant, uninterrupted availability of all computational and networking resources; latency must also be kept as low as possible.
2. Storage capacity
An increasing mix of cameras, LIDARs, radars, motion sensors is used to test and implement complex ADAS and driver assistance capabilities. These data are often stored at full resolution to perform in-vehicle, real-time operations (AI inference or reinforcement training) or offline AI training or simulations in large lab workstations. In-vehicle data loggers need to provide flexible storage capacity that is orders of magnitude larger than the typical size associated with traditional embedded computing devices.
3. Very high bandwidth
As shown in the previous section, an autonomous vehicle generates a continuous stream of data of about 4Gbit/s, which translates to 1.8TB/h in a simple case, and can go up to 40Gbit/s, or 18TB/h in a richer scenario; therefore, autonomous driving imposes a data logging speed to keep up with sensor feeds.
4. Ruggedness & automotive certifications
It is mandatory that high-performance computers can operate reliably when installed in a moving vehicle and while being subjected to the very harsh conditions typical of automotive applications. This requirement results in a dilemma: high-performance components are designed for highly controlled and benign environments, while rugged devices sacrifice performance to reduce MTBF (Mean Time Between Failure) by avoiding fans, vents, and other potential failure points. Since it is very hard to meet the necessary level of performance and reliability, certifications play a fundamental role in establishing an objective and measurable level of fitness of the system for the specific applications.
Automotive certifications, such as E-Mark, ECE ONU R10, ISO 16750 and IEC 60068-2-6 / 60068-2-27 are objective ways for characterizing the behaviour of the system under stress in actual operating conditions.
The physical dimensions of the systems need to be as small as possible, to allow the installation of the edge computers and data loggers in the vehicle; the limitations of air cooling impose compromises and hard choices. A better approach is to switch to liquid cooling, a technology that allows for rugged and compact design. Since liquid cooling is already part of almost every vehicle, including full-electric ones, it is also a natural choice that reuses part of the existing infrastructure and, at the same time, it makes much more efficient use of power.
We leverged our extensive experience in rugged computing and HPC to create a complete portfolio of high-performance systems that can be employed to create “data centers on the wheels”, i.e. rugged and certified data loggers and edge computing devices that bring capabilities and performance typical of server environments into a vehicle trunk to enable ADAS and, in the future, fully autonomous vehicles.