In his keynote speech, Nvidia CEO Jen Hsung-Huang went to great lengths to convince the auditorium that AI is the better approach than normal software coding for a broad range of applications including autonomous driving. Highlighting the performance progress in the area of neural networks (a specific approach within AI) through GPU computing, Hsung-Huang said that this AI branch is now at the verge of widespread industrial deployment. “We are able to solve real problems now – and I mean large ones”, he said. For autonomous driving, AI will increase safety by a large margin. “It is not conceivable we could achieve this level of safety with traditional image recognition.”
The Nvidia CEO and Co-founder also highlighted the Drive PX2 single-board computer as a platform to establish AI in the car and act as an enabler for autonomous driving. The platform is scalable and will be available in three versions - the smallest one is the “Autocruise” version, designed for autonomous driving situations in less critical environments such as highways with relatively constant speeds. The next higher model, the “AutoChauffeur” features more computing power with two instead of one GPU whereas the largest one will be able to support full autonomy for driverless cars. In terms of functionality, the Drive PX 2 can be used for the perception, reasoning and driving parts of the Autonomous Driving decision scheme (which means that Nvidia has at least the ambition to cover them all).
The necessary software can be created with the Driveworks Alpha 1 SDK that has been available for a while, but Hsung-Huang announced a new version with additional functionalities, including support for free automotive AI applications like space detection, distance detection and lane detection. The new release will be available coming October.
Hsung-Huang also announced a new System-on-Chip: Xavier. This SoC of which however the availability date is yet unknown will offer an AI computing performance comparable with a