Photo: David Herrera, CC BY 2.0
Highway technology reporting has a structural bias, and it is not political. A demonstration produces a date, a photograph and a quotable figure. An evaluation produces a confidence interval, a caveat about the comparison sites and a benefit-cost range that sometimes includes numbers below one. The first is easy to publish and the second is what an agency needs before it commits a capital program, so the coverage of this field runs years ahead of its evidence.
This section is organised around that gap. Every page here sorts its subject into what is deployed at scale with published results, what exists as a funded pilot, and what exists as an announcement, then says which is which. The sorting matters most where the three are routinely blended, which is why the pages on smart highways, artificial intelligence and digital twins are the longest.
The measurement standard the field mostly ignores
The strongest evidence in American traffic operations came from an experiment nobody would be allowed to run today. In 2000 Minnesota switched off every ramp meter in the Twin Cities for six weeks under a legislative order and measured what changed. Crashes, travel times, speeds and volumes all moved, and the direction of movement settled an argument that modelling had not.
Almost nothing since has been tested that way. The future of smart highways sets the Minnesota result against what has been published on the technologies that followed: a federal evaluation of variable speed limits whose three study corridors disagreed, and the integrated corridor management demonstrations where the before-and-after field analysis had to be abandoned for want of activations, leaving a modelled benefit-cost range that straddles break-even at one of the two sites. That is what honest evaluation of a partially deployed system looks like, and it reads very differently from the launch coverage of the same projects.
The same page prices the frontier by the quarter mile, using Indiana’s wireless charging testbed, where an electric truck drew 190 kilowatts at highway speed over a quarter mile of instrumented concrete. The unit of account there is energised lane miles, not stations built.
Where the software actually sits
Artificial intelligence in this sector lives mostly in the vans that survey pavement rather than in traffic management centres, and how AI is used in transportation infrastructure follows it there, because that is where a model’s output becomes a number in a pavement management system and then a project in a program. The National Academies survey behind that page found agencies unable to say which models process their own data, unable to confirm compliance with the relevant automated cracking practice, and overwhelmingly reliant on a handful of vendors. Five of twenty-seven had checked the machine results against a manual survey.
The most quoted success in the field, the Pittsburgh adaptive signal pilot, turns out to rest on twenty-four timed drive-throughs of nine intersections in 2012. The paper is candid about that. Its percentages travel without it.
Digital twins for highways and bridges takes the terminology problem head on, by setting the National Academies definition of a digital twin beside the definition FHWA’s own resource centre uses. They do not match. One requires a bidirectional link to the physical asset and a predictive capability; the other describes a geospatially located asset record. Minneapolis has been collecting continuous structural data from a major crossing since the bridge was rebuilt, without ever claiming the word.
The two deadlines
Charging and connectivity both move fast enough that every figure here needs a date attached, and both turn on federal decisions rather than on engineering.
How EV charging will change interstate highways separates the regulation, which still requires four fast charging ports at 150 kilowatts each and better than 97 percent annual uptime, from the guidance, which in August 2025 replaced a hard 50-mile spacing rule with a paragraph inviting states to consider what spacing suits them. The engineering floor held and the map became discretionary. Behind both sits the cost category the national investment estimate explicitly excludes, which is the grid connection.
Connected vehicles and V2X has a date on the calendar: licenses for the older 5.9 gigahertz radio standard cannot be renewed past 14 December 2026, which affects roadside equipment that states installed under federal encouragement. The page traces the chain from a 1995 crash investigation in fog, through the 1999 allocation of 75 megahertz, to three announced manufacturer deployments that never happened and a proposed federal mandate withdrawn in 2023.
What runs today, and what it cannot tell you
Underneath all of it is the unglamorous estate described in what intelligent transportation systems actually do: signs, sensors, cameras, weather stations and the staff who watch them. That page carries the most uncomfortable finding in the section. The 20 percent secondary-crash figure that has justified incident management spending for two decades sits alongside a 2023 federal study of ten states in which under 1 percent of crashes were flagged as secondary and only about 30 percent of those could be verified.
The live question across every page here is the same, and it concerns denominators rather than capability. These systems are justified by events that did not occur, measured against exposure that no agency counts. Until that changes, the field will keep choosing between demonstrations it can photograph and evaluations it cannot finish.