Last year, Nvidia announced their new Xavier AI Car Supercomputer. Due to the computing demands of an AI car, a computer would be required to perform the complex task of autonomous driving. While the hardware is there, the software side has remained mostly a mystery till today. During their CES keynote, Nvidia revealed their new AI Car Platform.
With 8-core custom ARM64, 512 core Volta, 30 TOPs DL, and 30W TDP, the computer runs the Driveworks OS for ASIL D Functional Safety. While the Xavier is the same as originally launched, the TOPS (terra operations) have been improved by 50%. To showcase their Xavier and the new platform, Nvidia has built the BB8 test AI vehicle which is more or less fully autonomous. The system is even capable enough to determine when it cannot handle certain parts of a trip (ie bad maps or too complex) and hand control back to the driver. The driver can also take back control whenever they choose to.
In addition tot he AI pilot, Nvidia is also introducing an AI Co-Pilot. This works as an assistant, warning the driver about various obstacles and hazards it picks up from sensors. It is also capable of facial recognition, head tracking, gaze tracking and lip reading. This allows the co-pilot to caution the driver if they are perhaps in the best state to drive (drunk or distracted or emotional) and warn about specific dangers the driver can’t notice due to their positioning.
Fully integrating all of those functions, of course, is the entire Nvidia AI Car Platform. The AI is also capable of handling natural language instructions as well as providing natural language feedback to the driver. There is also a connection to the cloud platform with mapping data being live streamed both ways. Even though the designs are still in the earlier stages, Nvidia is already partnering with ZF, Bosch, and Audi with the first real cars on the street with the system in 2020. I for one can’t wait for my car to drive me around for once.
