The U.S. Department of Transportation is paying three teams to build AI systems that predict crashes at intersections, backed by $5.5 million in Phase 1 funding. A Detroit company called Derq is one of them.
The company announced the award in a Business Wire press release on Thursday. The program also asks for research on whether broader deployment pays for itself. That cost-benefit answer decides if your city ever buys this stuff. Cities already weigh similar trade-offs with automation elsewhere, as in the debate over Amazon’s million robots.

What USDOT picked for its intersection safety system
Derq says it was chosen for Phase 1 of USDOT’s Intersection Safety Systems program, known as ISS. Per the release, the other two teams are UCLA and the University of Michigan Transportation Research Institute. That makes Derq the only private company in the trio.
The job is to build and demonstrate systems that spot dangerous moments between cars and pedestrians. They must also predict what happens next and step in before someone gets hurt. Think of it as a crash prevention system for the corner itself. Derq says it will test with transportation agencies and partners in Texas and Virginia.
The $5.5 million is split across all three teams. Derq’s own share wasn’t disclosed, so don’t picture a $5.5 million check landing in Detroit. The department also anticipates another $11.5 million for Phase 2, which would test refined prototypes at real intersections. USDOT’s own page describes the opportunity as up to $17 million across three teams. Phase 2 money isn’t guaranteed, and nothing here says every team moves on.
Why a smart intersection matters
Intersections are where cars, bikes and people all try to occupy the same spot at the same time. The release cites roughly 36,640 U.S. traffic deaths in 2025 and says intersections account for about 18% of pedestrian fatalities. Other USDOT-linked material puts the intersection share of all traffic deaths closer to a quarter.
Here’s the pitch. Derq says its system blends camera video, other sensor feeds, traffic-signal data and vehicle-to-everything messages, then crunches it at the roadside. CEO Georges Aoude puts it this way: “Most systems today tell you what happened. Ours also predicts what’s about to happen and acts on it.”
That kind of AI traffic safety pitch is hardly unique to roads. The same debate over how much to trust machine judgment shows up in the Pentagon’s fight over AI ethics.
This didn’t come out of nowhere. USDOT ran an Intersection Safety Challenge from April 2023 to January 2025. It drew 120 concept papers in its first round and handed out millions in prizes. Derq took one of the $750,000 top-tier awards in the second round, alongside UCLA, the University of Hawaii and the University of Michigan.
Those results showed many teams could detect, locate and classify road users under good conditions. The new program is about doing it in the messy real world, with rain, glare, odd angles and people who don’t read the manual. Demos in good weather are easy. That real-world part is the true test for any intersection safety system.
What happens next for the smart intersection push
If you’ve ever held your breath while a driver rolled through a crosswalk, this is the problem being worked on. But nothing at your corner changes tomorrow. Phase 1 is prototype building and testing, and USDOT hasn’t announced any national rollout.
Phase 1 testing is under way, and no Phase 2 date or decision has surfaced. Still open: which Texas and Virginia agencies are involved, and whether all three teams advance. I found no independent coverage yet, so treat the details as company-reported until USDOT says more.
For now, three teams have a few million dollars and a hard problem. Can an intersection really learn to flinch before the crash?






