Close
On Day 2 of the ongoing ITC India 2020 virtual conference, Riccardo Mariani, VP Industry Safety, Nvidia, spoke about “Architecting a software-defined car—from end to end and top to bottom”.
The market has the potential to grow to $700 billion. The software-defined architecture has opened up several areas. There is powerful and efficient AI, CV, AR, HPC, functional safety, etc. There is rich software development. An autonous driving platform has to be must. Nvidia has the 5W to 2,000 TOPS as one programmable architecture. The Nvidia Drive AGX Orin has 17 billion transistors.
We need a drive stack that is open and scalable. It creates the next generation of AR. It uses AI capabilities. AI gives functionalities such as perception, reasoning and driving. There is also HD map and mapping.
A video showed how the car will travel, especially at the intersections. The results can be used in several different ways to create additional diversity and redundancy.
To be complete, the autonomous driving architecture should be end-to-end. We are delivering access to the industry to the Nvidia DNN. We are accelerating the autonomous driving architecture. Validation by simulation is very important. Nvidia still has AV test fleets, it is impossible to cover everything for a self-driving car. The Nvidia drive scene, we can deliver self-driving technology for any scenario.
The video on Nvidia drive constellation showed the virtual maps, with added environmental details. You can configure the sensor array and create a variety of traffic scenarios. The day and weather conditions can also be modified and created. DRIVE SIM creates the virtual world.
Taking a deep dive into functional safety, we need to have cybersecurity, safety of intended functionality (SOFT), and functional safety (FUSA), that takes care of hardware failures and systematic failures. There is the ISO/SAE 21434 for road vehicles cybersecurity engineering, ISO 21447 for road vehicles safety of intended functionality, and ISO 26262 for functional safety.
There is also the ISO/TR 4804 for safety and security for ADAS design, verification and validation methods. The UL 4600 is for the evaluation of autonomous products. The ISO/IEC JTC1 SC 42 WG3 is also there. The IEEE P2846 is formal model for safety considerations. IEEE P1228 is for software safety. The IEEE P2851 is for exchange/interoperability format for functional safety analysis and verification. The hardware architecture has primary channel and backup channel.
The system-level monitors are trajectory and command monitors, and collision avoidance. There are function-level monitors, such as output validity and cross checks, and valid output trends. There are low-level diagnostics such as hardware diagnostics and self-checks, and program flow monitoring. Detecting a fault is not enough, and emergency operation is also critical. The system emergency operation is also necessary.
DNN functional safety is built in. The data factory, planner, DNN V&V, and tools and AI infrastructure are necessary. The autonomous driving architecture should be software defined, AI-enabled, and be functional safe.
Earlier, Dr. Yervant Zorian, Chief Architect and Fellow, Synopsys, President, Synopsys Armenia, presented on ITC’s activities. He said that at the global ITC, ITC India participates actively. We have tried to broadcast all plenary sessions also, before the pandemic. It is beneficial to have a virtual presence. We are investigating virtual conference tools. We have the ITC virtual conference meeting on July 21st, 2020. We will be holding three special tracks on automotive, security, and AI. Together, all the contributions are also making ITC India powerful.
VP Industry Safety, Nvidia
Chief Architect and Fellow, Synopsys, President, Synopsys Armenia