SAE levels of Automated Driving range from 0 to 5, with Level 5 representing full automation and monitoring of the driving environment — under all roadway and environmental conditions. A Tech Briefs reader asked our expert: What is the best data-processing architecture for truly autonomous, "Level-5" vehicles?

An Autonomous Driving, or AD, application requires the ability to process large amounts of data. In addition, AD often calls for Deep Learning algorithms – mullti-layer neural networks that are “trained” to understand a given set of parameters.

Not to be confused with ADAS, or Advanced Driver Assistance Systems, the AD scenarios require significantly more processing power and a greater number of safety certifications.

With Automated Driving, cars “perceive” the road via cameras, lidar, and radar sensors. The data is then captured and processed.

In the data-streaming, or “planning” phase, the car’s central processing unit (CPU) handles information from the sensors so that a car can make a decision — determining a steering angle, for example.

One way to process the data is to have a smaller CPU combined with two bigger, more powerful graphics processing units (GPUs). The perception data goes first to the CPU, then to the GPU memory, and then back to the CPU for a final decision.

Another option: Two field programmable gate arrays (FPGAs), or hardware-configurable integrated circuits, can receive the camera’s and radar’s data directly. The data flow is then fed to the CPU that communicates the result to the car.

“This option offers the best balance between data transfer and processing," said Mohamed Bergach, System and Software Architect at the Augsburg, Germany-based embedding computing manufacturer Kontron.

In a live presentation titled Reconfigurable Chips for Automated/Connected and Cyber-Secure Vehicles, a Tech Briefs attendee asked Bergach: “What is the best architecture that will help to achieve Level 5 autonomous driving systems?”

Bergach had the following response:

“Level 5 autonomy is very stringent, and in my opinion, it will require a balanced architecture, with a mix of CPU and FPGA that allow a hybrid workload to be processed efficiently, and also will be able to comply with the temperature and power consumption requirements that are typical on ADAS systems. Both the CPU and FPGA can be the solution in the near future.”

What do you think? What technologies will support Level-5 autonomous driving systems? Share your thoughts below.

Watch the full presentation: Reconfigurable Chips for Automated/Connected and Cyber-Secure Vehicles