InfrastructureIssue 0213 records
Transportation
This cycle's transportation AI evidence is about detection and bounded pilots. Caltrans District 5, a VDOT-sponsored wildlife warning pilot and RMIT's pothole sensing report detection accuracy or counts, not safety outcomes, while Rekor's FDOT and Southern Nevada contracts show incident detection bought through ceilings and options. Bus and signal AI gains are real but corridor-specific: 30 percent in Belo Horizonte, 7 to 9.8 percent on two LA Metro routes and 1.5 percent in London. APTA's survey shows few transit agencies in production. Airports are testing AI assistants, robot parking and robotaxi curbs within explicit limits.
- Evidence records
- 13
- Cross-source patterns
- 4
- Evidence classes
- 6 independent reporting2 government evaluation2 academic research2 vendor claim1 public-sector association guidance
- Outcomes
- 8 emerging3 effective2 mixed
- Source freshness
- 10 new this fortnight2 recent1 undated
- Research completed
- 2026-10-31
Choose a role to see its takeaway beside every record in the ledger.
Patterns across the evidence
Synthesis · Lighthouse Advisory interpretation
Roadway AI detection reports accuracy and counts, not safety outcomes
Caltrans District 5 reports 95 percent or greater counting accuracy, the VDOT-sponsored wildlife pilot 97 percent detection and a 3 mph average slowdown, RMIT more than 300 recorded defects, and Rekor intended faster incident response. None of the four reports crash, injury or maintenance-cost outcomes, and most give no accuracy method.
Operating questionBefore scaling a detection system, which outcome metric (crashes, response time, repair time) will the agency measure against a baseline, and how will it publish the accuracy method?
Supporting evidenceCalifornia Transportation Commission; California Department of Transportation (Caltrans) District 5Virginia Tech Transportation Institute (VTTI); Virginia Department of Transportation (sponsor)RMIT University; RACVRekor Systems (vendor); Florida Department of Transportation; Regional Transportation Commission of Southern Nevada
Measured bus and signal AI gains are corridor-specific and often modest
Belo Horizonte reports 30 percent faster buses on a nine-intersection corridor but its own official says results will vary; LADOT reports 7 to 9.8 percent speed and reliability gains on two camera-enforced routes; CitySwift reports 1.5 percent shorter journeys in London; Spruce Grove expanded adaptive signals with no published metrics.
Operating questionWhat corridor-level before-and-after measurement, with control routes where possible, will each expansion phase publish?
Supporting evidencePrefeitura de Belo Horizonte (municipal mobility subsecretariat; Prodabel)Los Angeles City Council; Los Angeles Department of Transportation (LADOT); LA MetroCitySwift (vendor); Transport for London and London bus operatorsCity of Spruce Grove
Transit AI is pilot-heavy and bounded by data access and oversight
APTA's survey found five or fewer agencies per category actually using AI tools, citing data silos; Transit Tech Lab's eight-week proofs of concept convert to scale at a low rate (23 of 126); Los Angeles approved bus-camera expansion only alongside an ALPR policy and amid concerns about contractor data access.
Operating questionWhich data-access agreements and oversight steps must be in place before a transit AI proof of concept becomes a production pilot?
Supporting evidenceASCE Civil Engineering Source, reporting on the American Public Transportation Association AI/ML primerTransit Tech Lab (MTA and Partnership Fund for New York City), with Port Authority of New York and New JerseyLos Angeles City Council; Los Angeles Department of Transportation (LADOT); LA Metro
Airports are testing customer-facing AI and automation within explicit bounds
The Port Authority scoped AI assistants per airport on curated, staff-maintained content; Korea Airports Corporation limited robot valet parking to daytime hours and eligible vehicles; TRB webinar speakers described airports planning curb strategies for robotaxis now serving six U.S. airports. None reports performance data yet.
Operating questionWhat accuracy, incident and customer metrics will the airport collect during bounded pilots before widening scope?
Supporting evidencePort Authority of New York and New JerseyKorea Airports CorporationGovernment Technology, reporting on a Transportation Research Board webinar
Evidence ledger
Full record · every source keeps its link and limitations
Innovations in Transportation: Artificial Intelligence Cameras in Caltrans District 5 (CTC information item 4.4)
Caltrans District 5 piloted an AI camera on State Route 154 from July 2023 that detected, classified and counted most roadway users (pedestrians, bicyclists, cars, trucks, buses, motorcycles) with a stated accuracy of 95 percent or greater. A SHOPP Minor B project then installed 10 more AI cameras in February 2025 at high pedestrian and bicyclist locations, providing count data not previously available. The district is exploring making AI cameras a standard feature of future camera and signal installations; the cameras do not read plates or collect PII.
Why it matters, evidence and limitations
- Why it matters
- A state DOT moving a narrow, privacy-limited AI use (multimodal counting) from pilot to funded capital deployment and possible design standard is a transferable path for other DOTs' active-transportation data programs.
- Evidence and measured results
- Agency memorandum summarizing a commission presentation. The 95 percent accuracy figure has no stated method, sample size, ground-truth process or per-class breakdown; cost and staff-time benefits are asserted, not quantified.
- Limitations and uncertainty
- Self-reported by the agency; single-district experience; no accuracy methodology or per-mode error rates; benefits are described qualitatively.
Elk crossing: Real-time driver warning system helps prevent animal collisions (VTTI pilot sponsored by VDOT)
A one-year VDOT-sponsored pilot on Route 460 used thermal cameras and AI software (from vendor Crosstek) on eight portable, solar-powered trailers with dynamic message signs over a 750-meter section. The detection system showed 97 percent overall accuracy for medium and large wildlife, and drivers slowed by an average of 3 mph approaching activated signs, based on more than 10,000 passing vehicles measured with roadside radar. The pilot ended in September 2026 and analysis is continuing.
Why it matters, evidence and limitations
- Why it matters
- Shows a portable AI hazard-detection and warning pattern a state DOT can deploy in rural, low-connectivity corridors, with a measured driver-behavior response rather than only detection accuracy.
- Evidence and measured results
- University news release on a research pilot. Speed change is measured by radar across 10,000+ vehicles; the release does not report crash outcomes, false-alarm rates, or the accuracy evaluation method. Analysis is ongoing, so results are preliminary.
- Limitations and uncertainty
- Preliminary; one corridor with roughly 100 elk; a 3 mph average reduction is modest and crash reduction is not demonstrated; vendor partnership and an award are mentioned in a promotional context.
Projeto de semáforos com IA aumenta em 30% velocidade dos ônibus em BH (AI bus signal priority pilot, Belo Horizonte)
Reporting on a city briefing to the municipal council on 22 October: a pilot since May uses cameras and AI to detect approaching buses and adjust signal timing at nine intersections on two streets. The city reports bus speeds rose 30 percent and morning-peak travel time fell about three minutes on the corridor, with about 8,874 daily passengers benefiting and no significant congestion increase on cross streets. The city plans 576 controllers by December and completion in 2027, integrated with its operations center. An official cautioned results will not be the same citywide; councillors questioned site selection, measurement, and whether the system could be repurposed for public-security monitoring.
Why it matters, evidence and limitations
- Why it matters
- A measured transit signal priority result plus a rapid scale-up plan offers U.S. agencies a comparison point, and the council debate shows the scrutiny questions (measurement, repurposing) that expansion invites.
- Evidence and measured results
- Press report of municipal figures presented to council; no methodology, baseline period, or independent evaluation is described. Transferability to U.S. signal systems and transit operations is limited.
- Limitations and uncertainty
- Single short corridor of nine intersections; city-reported figures; the official explicitly said gains will vary by location; reporting is in Portuguese and from a Brazilian context.
Transit Tech Lab Releases Proof-of-Concept Results Addressing Infrastructure and Workflow Modernization
Seventeen companies ran eight-week proofs of concept with the MTA and Port Authority across infrastructure and data/workflow challenges. AI items include generative drafting of 15 LIRR technical scopes of work with a reported 40 percent drafting-time cut, an AI insight engine giving technicians guidance from manuals and service records on a street sweeper, AI-assisted facility and asset management at JFK's Bergen Substation and the Bayonne Bridge, aerial mapping of Port Authority airport and marine terminals, AI curb and bus-terminal tracking scoped for the Midtown Bus Terminal, and AI voice deepfake detection. The lab reports 126 proofs of concept, 64 pilots and 23 commercially scaled technologies to date.
Why it matters, evidence and limitations
- Why it matters
- Gives transit and port agencies a broad, recent menu of AI knowledge-work and asset-management proofs of concept and a funnel metric (126 POCs to 23 scaled) that sets realistic expectations for conversion.
- Evidence and measured results
- Program summary of company-run proofs of concept; most items describe activity rather than measured results. The 40 percent drafting-time figure has no stated method. The related Ontra estimates were already archived and are not re-reported here.
- Limitations and uncertainty
- Eight-week, company-led tests; results are summaries without methods; inclusion does not mean agency adoption; promotional quotes from program partners.
Transit agencies are testing AI's potential, report finds, with broader adoption likely
Reporting on APTA's May 2026 primer, which surveyed 32 transit agencies: experimentation or consideration is highest in back office (50 percent), operations (47), customer support (44) and maintenance (41), but in every category only five or fewer surveyed agencies actually use AI tools. The article cites MTA camera enforcement results reported in the primer (bus lane speeds up 5 percent, collisions down 20 percent on average on relevant routes), a $60,000 CTA rail right-of-way intrusion pilot that reached 99 percent accuracy per the contractor, and researchers who say adoption is limited by data silos rather than AI. CapMetro cautions that oversight, pricing models and process quality determine whether uses are sustainable.
Why it matters, evidence and limitations
- Why it matters
- Newly relevant because it puts APTA's adoption data and practitioner cautions in front of the engineering audience this month; it supplies the base rate (few agencies in production) against which this week's transit AI announcements should be read.
- Evidence and measured results
- Secondary reporting on an association survey of 32 agencies (not a random sample) plus interviews; the 99 percent CTA figure is a contractor statement; MTA figures are quoted from the primer without method.
- Limitations and uncertainty
- APTA primer was published in May; survey size is small; several quantitative claims are secondhand; the article includes event promotion.
Port Authority wins award for airport websites featuring new AI assistants
The Port Authority's redesigned websites for JFK, Newark, LaGuardia and Stewart won ACI-NA's 2026 best airport website award. Each site has a purpose-built AI assistant developed and trained by Port Authority staff, unique to that airport and built on a curated knowledge base maintained by subject matter experts. The assistants pull live operational data such as flight information and wait times and draw on verified sources including the Port Authority site, DHS, TSA, CBP and CLEAR. Larger airports add experience planners that can suggest accessible routes based on stated mobility needs.
Why it matters, evidence and limitations
- Why it matters
- A large public airport operator built in-house, airport-specific assistants grounded in curated and federal sources, a pattern other airports and transit agencies can compare with vendor chatbots.
- Evidence and measured results
- Republished agency announcement of an industry award judged on design, communications, accessibility and navigation; no accuracy, usage or customer-satisfaction metrics for the AI assistants are given.
- Limitations and uncertainty
- Award recognizes the websites overall, not AI accuracy; source is a republished press announcement; launch was months earlier and no performance data is provided.
Gimpo Airport Launches 'AI Robot Valet Parking' Service at International Terminal
Korea Airports Corporation, a state-owned airport operator, began piloting an 'AI Robot Valet Parking' service in Gimpo's international-terminal underground parking lot: passengers leave vehicles in a designated area and robots lift and move them to spaces. Operations are limited to 10 a.m. to 5 p.m., with nighttime excluded to ensure stability of the initial rollout; the service is free during the pilot, requires reservations with mobile identity verification, and excludes vehicles over set size and weight limits. The operator will survey satisfaction and may expand to other airports.
Why it matters, evidence and limitations
- Why it matters
- A public airport operator testing robotic parking with explicit operating limits shows how automation pilots can be bounded by hours, vehicle eligibility and reservations; relevant to U.S. airport parking modernization as a comparison only.
- Evidence and measured results
- News report of an operator announcement on launch day; no performance, throughput, incident or capacity data. The page does not explain what the AI component does beyond the service name.
- Limitations and uncertainty
- Launch announcement only; AI functions not specified; Korean regulatory and operator context; vehicle size limits exclude some large sedans and SUVs.
Los Angeles expands bus lane enforcement with DASH bus cameras
The City Council voted 13-1 to allow up to 50 cameras on DASH buses and approved LADOT's updated ALPR use policy, with rollout from Nov. 1 on 30 buses and 60 days of warnings, via a three-year, $3.43 million piggyback on Metro's Hayden AI contract. A 60-day pilot logged more than 2,400 violations (about four per bus per hour). LADOT reports Metro's program (Nov 2024-Jan 2026) sent 281,000 potential violations, generated over $21 million gross, improved speed and reliability on routes 212 and 720 by 7 to 9.8 percent, and saw about a 33 percent drop in citations after a year. Councilmembers raised concerns about contractor access to data and allegations involving the vendor's founders; the program depends on state authority that expires Jan. 1, 2027 unless AB 1837 is signed.
Why it matters, evidence and limitations
- Why it matters
- Directly informs agencies expanding AI camera enforcement: measured route-speed gains coexist with live council concerns about vendor data access and legal authority, and with an ALPR policy as a precondition.
- Evidence and measured results
- News report of a council vote with LADOT-reported metrics; the figures are agency-reported without method, and the article does not report accuracy or false-citation rates.
- Limitations and uncertainty
- Agency-reported outcomes; no accuracy or false-positive data; vendor allegations are stated by a councilmember and not adjudicated in the article; program continuity depends on state legislation.
FDOT Increases Rekor Contract Ceiling by 94% to $9.5 Million; RTC of Southern Nevada Selects Rekor Command
Rekor says FDOT raised the ceiling of its statewide intelligent transportation infrastructure contract from $4.9 million to $9.5 million, but states that the ceiling is not committed revenue: $4.9 million is approved and encumbered and further use depends on funding and task orders. RTC of Southern Nevada selected Rekor Command for traffic management across Clark County with a potential value of about $1.54 million including two option years; the deployment integrates AI incident detection and prioritization, driver alerts, public safety dispatch data, and NDOT's traffic management system, intended to shorten incident response and reduce secondary crashes.
Why it matters, evidence and limitations
- Why it matters
- Shows how AI incident detection is being bought by state DOTs and regional agencies through ceiling increases and option-based contracts, and why contract values in vendor releases need careful reading.
- Evidence and measured results
- Vendor press release with forward-looking-statement disclaimers; no performance data for incident detection speed or accuracy; agency confirmation is not included.
- Limitations and uncertainty
- Vendor-authored; values are ceilings and potential totals; outcomes are intended, not measured; the company discloses funding and integration risks.
'Pothole patrol' pilot hits the road with AI-powered road defect monitoring
RMIT equipped an RACV roadside-assistance patrol van with RoughSense, combining AI, GPS, video and vehicle-movement data to detect potholes, cracking and rutting in near real time. Between 21 August and 21 October 2026 the system recorded more than 300 defects across about 1,000 km of roads on the Mornington Peninsula, with repeat passes showing worsening potholes. The patrol driver manually flags defects to validate detections.
Why it matters, evidence and limitations
- Why it matters
- Fleet vehicles already driving public roads can double as defect sensors; road agencies can compare this with municipal camera-based programs, though this pilot is a research trial with a motoring club, not a road authority.
- Evidence and measured results
- University news release; counts of recorded defects, not accuracy, precision or recall against inspections; manual validation is described but results are not reported.
- Limitations and uncertainty
- Two-month trial with one vehicle; no accuracy metrics; not a road-authority deployment; Australian context.
Airports Can Serve as Hubs for Transportation Innovation
At an Oct. 20 TRB webinar, researchers and planners said airports are testbeds for autonomous vehicles and robots. Six airports are now served by Waymo robotaxis (San Francisco, San Jose, San Antonio, Phoenix, Las Vegas, Nashville). Researchers outlined three curbside strategies for robotaxis: dedicated curb, shared curb, or remote staging with call-up. National Lab research for Dallas Fort Worth aimed to set planning time frames for adopting new mobility. Airports are also considering autonomous people movers and robotic baggage and cargo handling.
Why it matters, evidence and limitations
- Why it matters
- Public airport operators must plan curb space, staging and coordination for AI-driven vehicles now that robotaxis serve multiple U.S. airports.
- Evidence and measured results
- Reporting on webinar remarks; no performance or safety data for airport robotaxi operations; the curbside strategies are planning options, not evaluated outcomes.
- Limitations and uncertainty
- Planning discussion, not evidence of results; speakers include consultants; limited detail on the DFW study.
NoTraffic adaptive signal technology (City of Spruce Grove)
After a two-intersection pilot that began in April 2025, the city expanded AI-based adaptive signal control from NoTraffic to 12 intersections along Highway 16A and Jennifer Heil Way, fully operational the week of October 13, 2026, following left-turn operation changes on September 28. The city states the system does not collect personal information, read plates or issue tickets, that vehicle speed data is not stored or linked to identifying information, and that it remains responsible for the system and will monitor performance. No performance results are published.
Why it matters, evidence and limitations
- Why it matters
- A small city's public explanation of an AI signal system's data limits and accountability is a practical template for U.S. municipalities, though it publishes no outcome data.
- Evidence and measured results
- City information page; describes deployment and data practices, but provides no measured delay, travel-time or safety results from the pilot or expansion.
- Limitations and uncertainty
- No published metrics; the page calls the pilot successful without data; publication date not stated; Canadian municipal context.
Bus passengers save 1.5% in journey time on London routes optimised by CitySwift's AI
CitySwift reports that customer journey times on routes optimised with its AI platform fell 1.5 percent, measured using TfL's Journey Time Metric combining speed and reliability. It says this equals 27 seconds per optimised journey, projects more than 25 million passenger-hours saved over three years valued at over £230 million, and states more than 40 percent of London's bus network has been reviewed with 200 routes improved at no cost since deployment in 2024.
Why it matters, evidence and limitations
- Why it matters
- A measured, modest result using an agency-defined metric is a useful counterweight to larger projections in transit AI marketing; U.S. agencies can ask vendors for similar agency-metric reporting.
- Evidence and measured results
- Vendor announcement; the metric is TfL's, but the analysis, baseline and attribution method are not published; the hours and value figures are projections.
- Limitations and uncertainty
- Vendor-authored; no independent evaluation; projections of hours and monetary value; London's franchised bus model differs from most U.S. agencies.
How to read this edition
Source findings, measured results and limitations come from the cited publications. Patterns, operating questions, role takeaways and implementation considerations are Lighthouse Advisory interpretation, stated as questions to validate locally rather than guaranteed outcomes. Vendor and operator claims are labeled as claims. Full research method.
- Independent reporting
- Independent reporting with attributable sources but without a formal evaluation design.
- Government evaluation
- A public body’s measured evaluation or documented pilot.
- Academic research
- Research produced through an academic institution or peer-reviewed venue.
- Vendor claim
- A supplier-provided assertion that has not been upgraded to independent evidence.
- Public-sector association guidance
- Practitioner guidance or an association-supplied case; not independent outcome evidence.