Nvidia says its new AI autonomous driving platform offers the redundancy needed to enable the development of functionally safe self-driving cars and trucks that can meet safety standards such as ISO 26262. The platform is designed to allow safe operation even in the face of faults related to the operator, systems or environment.
To achieve fail operation, Nvidia Drive can incorporate various processors including Nvidia-designed IP related to Nvidia Xavier covering CPU and GPU processors, image processing ISP, computer vision PVA, deep learning accelerator and video processors. It also incorporates lockstep processing and error-correcting code on memory and buses, with built-in testing. The highest system ASIL-D rating can be reached with the ASIL-C Nvidia Drive Xavier processor and ASIL-D-rated safety microcontroller, Nvidia says.
In terms of the Nvidia Drive OS software, key partners include BlackBerry QNX for its 64bit real-time operating system and TTTech for its MotionWise safety application framework, which isolates applications and offers real-time computing. The software supports the Adaptive Autosar open-standard automotive system architecture and application framework. Nvidia’s toolchain incorporates the CUDA compiler and TensorRT, and uses ISO 26262 tool classification levels.
The Nvidia Drive AV software stack is capable of functions including ego-motion, perception, localisation and path planning each with a strategy for redundancy and diversity, for example sensor fusion creates redundancy in perception. Redundancy and diversity are also optimised through deep learning and computer vision algorithms running on CPU, CUDA GPU, DLA and PVA. The result acts as a backup system to the car maker’s self-driving stack, enabling functionally safe Level 5 automation, says Nvidia.
The Nvidia AutoSIM virtual reality simulator, running on Nvidia DGX supercomputers, enables the platform to be tested for rare conditions.
