USPS considering 30 edge AI applications to automate mail processing

The new capabilities would boost the new Edge Computing Infrastructure Program's ability to process mail.
A USPS mail worker casts a shadow while wheeling boxes near Japanese Cherry Blossom trees in the West Village, amid the coronavirus pandemic on April 07, 2021 in New York City. (Photo by Alexi Rosenfeld/Getty Images)

The U.S. Postal Service is considering about 30 artificial intelligence applications for the Edge Computing Infrastructure Program (ECIP) developed in 2020.

Apps using optical character recognition (OCR) to streamline imaging workflow, automatically checking if packages have the right postage and deciphering damaged barcodes could all launch before summer.

In three weeks, USPS Senior Data Scientist Ryan Simpson and six NVIDIA architects designed deep-learning models capable of analyzing the billions of images generated by processing centers equipped with edge AI servers. Not only can the distributed edge AI system’s seven algorithms process 231 packages a second, but one is even capable of reverse image searching the 100 million packages USPS sees daily.

“Missing packages that used to take eight or 10 people several days can now be tracked down by one person in a couple of hours,” an NVIDIA spokesperson told FedScoop. “ECIP also enables rapid application deployment; new capabilities, that previously would have taken months, can be deployed in as little as two weeks.”


ECIP runs on the NVIDIA EGX platform across 195 USPS sites after most of the necessary hardware was finished in August, and already added a second computer vision app. NVIDIA’s Triton Interference Server functions like an automated digital mail person delivering AI and machine learning models when and how each of the sites needs them.

The forthcoming OCR use case will live as a deep-learning model in an ECIP container managed by Kubernetes and served by Triton, rather than requiring standalone IT infrastructure or a public cloud service.

“The models we have deployed so far help manage the mail and the Postal Service — it helps us maintain our mission,” said Todd Schimmel, manager of letter mail technology at USPS, in a statement.

But various USPS components ranging from enterprise analytics to finance and marketing have proposed the nearly 30 ECIP apps currently being considered.

AI apps require the real-time computing ECIP affords for large amounts of data.


“Because edge computing processes data locally, instead of in the cloud or a data center, it minimizes latency and bandwidth needs allowing for real-time feedback and decision-making,” said the spokesperson. “Edge AI solutions also provide greater security protections.”

ECIP and the apps being developed for the system play into USPS’s broader effort to make better use of the data it collects to both improve its efficiency and save taxpayer dollars.

Despite the progress USPS has made, edge AI remains a nascent technology.

“Every day, people in our organization are thinking of new ways to apply machine learning to new facets of robotics, data processing and image handling,” Schimmel said.

This story was featured in FedScoop Special Report: Automating Government Services - A FedScoop Special Report

Latest Podcasts