America’s freight system is a miracle of modern logistics, until it isn’t. One snowstorm, one labor shortage, one delayed truck outside a major hub, and the whole process starts to wobble. Packages miss delivery windows. Shelves sit empty. Costs spike. What’s exposed in those moments isn’t bad luck. It’s a system that operates like the world is predictable. In reality, freight moves through chaos.
That’s the problem Lacy Greening wants to solve.
Greening is an assistant professor of industrial engineering in the School of Computing and Augmented Intelligence, part of the Ira A. Fulton Schools of Engineering at Arizona State University. She was recently named one of 15 semifinalists in the U.S. Department of Transportation’s Advanced Research Projects Agency – Infrastructure, or ARPA-I, Innovation Challenge.
Chosen from 448 submissions nationwide, the recognition places her in a small, highly competitive cohort invited to pitch high-risk, high-reward ideas that could reshape how America’s infrastructure works.
Stuck in the middle mile
Her idea starts with a simple question: What if the U.S. freight system could become smarter and more resilient through networks that learn, coordinate and adapt in real time?
Right now, freight logistics don’t work that way. Planning systems are fractured. When something goes wrong, people scramble to stitch those systems together manually.
“Today, everything is planned kind of sequentially,” Greening says. “We do route planning, we do dock scheduling, we do sortation planning — all separately. And if there’s a delay in one place, it takes manual intervention to fix everything downstream.”
Her proposed solution uses agentic artificial intelligence, or AI. Instead of a single, massive optimization model trying to solve problems for the entire freight network at once, Greening envisions many smaller AI agents that reason locally but coordinate globally to connect those isolated planning systems. In practice, that means each part of the freight network can adjust in real time and communicate those changes to the rest of the system, without waiting for a human to step in.
“You can’t solve the whole problem at once. It’s way too big,” Greening says. “Right now, we have dedicated models. The problem is they don’t talk to each other. We want them to communicate without that manual handoff.”
The work is part of a broader collaboration with Reem Khir, an assistant professor at Purdue University. The team is zeroing in on the most and failure-prone and expensive part of the freight pipeline: the middle mile.
If the first mile moves goods from factories and ports into a company’s network, and the last mile delivers packages to front doors, the middle mile is everything in between. It’s the web of transfers between warehouses, fulfillment centers and regional hubs.
“Middle mile is typically the most complex part of the system,” Greening says. “There’s so much consolidation that has to happen, and that’s where a lot of the costs are tied up.”
It’s also where today’s tools struggle most. Middle-mile decisions are still largely reactive. When disruptions hit, manual responses can take 30 to 60 minutes or longer to roll out across a network. By then, delays have already cascaded.