The U.S. Postal Service often gets a bad rap for slow deliveries and budget overruns but lately has been embracing use of an edge artificial intelligence platform across 195 processing centers nationwide.
The AI benefits include the ability to track a missing package in a couple of hours instead of several days under a previous routine, said Todd Schimmel, a manager of letter technology at USPS, in a statement.
Training the USPS system for computer vision of packages was also vastly simplified with AI capabilities. What might have taken two weeks on a network of servers with 800 CPUs was reduced to 20 minutes on four Nvidia V100 Tensor Core GPUs in a single HPE Apollo 6500 server, Nvidia said.
Now, each edge server processes 20 terabytes of images a day from more than 1,000 mail processing machines. Nvidia relied on open-source Triton Inference Server to act as a digital mail carrier to the 195 processing centers when needed.
The issue with processing images of packages at the postal service is primarily one of scale. More than 7 billion packages are processed in a year, which comes down to 231 a second.
The USPS AI system came into being starting in 2019 when an AI architect Ryan Simpson began to tackle a method to share billions of images nationwide so that a missing package could be traced. Working with Nvidia and others, the team created the Edge Compute Infrastructure Program. USPS eventually issued bids for 10 data science personnel and servers, issuing contracts to Accenture Federal Services and HPE, according to Anthony Robbins, vice president for federal government clients at Nvidia.
Robbins said in a call with reporters that EPIC represents perhaps the largest AI computer vision project in the federal government and one of only a few in private enterprise.
“In 35 years of working with the federal government, I have not seen an agency act as rapidly in adopting a project as this one,” he said. “It’s a model for other government agencies and global entities.”
Other applications for EPIC will include using imaging to determine if a package is charged the right amount of postage for its size. Also, AI will be used to help decipher mangled bar codes.
Robbins estimated there are upwards of 1,000 AI proof of concept projects underway across the federal government heavily centered on armed services projects. Building out AI systems beyond the concept stage at a large scale is the next big challenge for both industry and government managers, he said.
“The innovation work underway is extraordinary,” he said. “AI projects in government and enterprise are as much about leading transformation and change as it is about the technology itself.”
RELATED: Nvidia CEO Jensen Huang and the future of…everything