AI in Traffic Engineering: What It Can Do Today — and What Stays With the Engineer

AI has moved from slideware to real tools in traffic engineering. But “AI” covers everything from image recognition to large language models, and not all of it belongs in a signed study. Here’s a practical map of where AI earns its place today, and where the engineer must stay firmly in charge.

Where AI already saves real time

Building the network from imagery

Coding lanes, turn bays, medians and traffic control by hand is slow and repetitive. Computer-vision models can now trace road edges, lane lines and turn arrows from aerial imagery and combine them with map data to build an analysis-ready network. The key is verification: the result should always be shown over the imagery so an engineer can check it at a glance and fix any approach in seconds.

Reading the files you already have

Counts arrive as spreadsheets and PDFs in a dozen formats; signal timing arrives as controller printouts from different vendors. Reading these into the model is tedious and error-prone. AI document reading can extract counts, timing plans, splits and time-of-day schedules and map them to the right approaches — with the engineer reviewing the result rather than retyping it.

Reasoning that’s written down

Trip distribution has always relied on judgement: where do the residents work, which route will they take? Language models are good at laying out that reasoning explicitly — “55% north on Main because the employment centres are north” — so a reviewer can follow it and an engineer can change it. Written-down reasoning is an improvement over percentages that live only in someone’s head.

Deterministic rules, applied consistently

Taper lengths, sign spacing, buffer distances, HCM delay calculations — these are formulas, not judgement calls. The value AI adds isn’t in the arithmetic; it’s in applying the right rule to the real geometry, everywhere, every time, and citing the source.

Plan review

Checking a plan set against MUTCD and a local supplement is a pattern-matching job at scale: every sign, every taper, every note on every sheet. AI review can catch the routine issues consistently and leave reviewers to focus on the design questions only a person can answer.

Reports and figures

Volume figures, LOS tables, study area maps and the narrative that ties them together are generated from the same data, so when a number changes, everything changes with it.

What must stay with the engineer

Scope and assumptions. Study area, analysis years, growth rates and which developments count as background are agreements with the agency, not outputs.
Judgement calls. When a driveway sits in a work zone or a distribution assumption is contested, the tool should ask, not guess.
Facts the model can’t know. Work hours, contractor contacts, a council’s preference — an honest tool leaves these blank instead of inventing them.
The seal. A licensed engineer reviews, edits, and takes responsibility for every deliverable.

What to look for in an AI tool

Traceability — every number links to its inputs and the standard applied.
Editability — every output can be changed, and changes flow downstream.
Benchmarks — the vendor can show results against real, published studies.
Honesty — it asks when it doesn’t know, rather than filling gaps with plausible guesses.
Data privacy — your projects stay in your workspace and aren’t sold or used to serve other customers.

How we approach it at EntentAI

LaneMarker AI Studio was built around those principles: AI Build is checked over the imagery, trip distribution explains every percentage, HCM results are benchmarked against published studies, and traffic control plans cite every rule and stop to ask when judgement is needed. The goal isn’t to replace traffic engineers — it’s to give them back the days they spend on formatting, so they can do more engineering.