For months, engineering teams from USPS and NVIDIA have been developing the AI models, which will help USPS read address labels more quickly and effectively. The system starts with high-performance servers powered by NVIDIA V100 Tensor Core GPUs and deep learning software, which train multiple algorithms. The trained models are then deployed to NVIDIA EGX edge computing systems at nearly 200 USPS facilities.

Delivery and testing of the system will begin this year, and the AI is expected to be fully operational in spring of 2020. While USPS will get an efficiency boost, this is a big win for NVIDIA too. It shows that AI processes are finding more practical uses.

USPS is experimenting with self-driving mail trucks as well. This spring, it signed a contract with San Diego self-driving truck company TuSimple for a two-week pilot. That involved the company’s trucks hauling USPS trailers on five round trips between the postal service’s Phoenix, Arizona and Dallas, Texas distribution centers. USPS is also developing a drone ‘airline’ to deliver medical samples in the US.