InfrastructureIssue 047 records
Transportation
This backfill edition finds transportation AI moving ahead of the evidence and authority needed to judge it. Incheon's airport runs AI safety, congestion and robot services, but its own coverage says benefits are unproven. FAA's SMART reports early savings as lawmakers seek a pause, and London's mayor wants robotaxi caps that TfL cannot impose. Lahore seeks citywide AI signal funding after an unreported pilot. DRJTBC's open-road tolling adds AI classification, Calgary extends apron AI to cargo, and Indonesia's KAI launches a rider chatbot that also collects behavior data. No item reports independently measured outcomes.
- Evidence records
- 7
- Cross-source patterns
- 2
- Evidence classes
- 5 independent reporting2 vendor claim
- Outcomes
- 6 emerging1 cautionary
- Source freshness
- 7 new this fortnight
- Research completed
- 2026-11-03
Choose a role to see its takeaway beside every record in the ledger.
Patterns across the evidence
Synthesis · Lighthouse Advisory interpretation
Airport and airspace AI is expanding faster than published performance evidence
Incheon runs AI airside safety, congestion forecasting and robots, yet its own coverage says accuracy and wait-time effects are unproven; FAA's SMART reports early DOT figures (110 rerouted flights, more than 10 minutes saved on average) without a stated method while national rollout is planned; Calgary goes live with cargo ApronAI on 10 November and plans to learn from peak season, with no gate-deployment results published.
Operating questionBefore an airport AI service is expanded or funded further, what baseline, metric and reporting owner has been agreed for it?
Supporting evidenceIncheon International Airport CorporationU.S. Department of Transportation / Federal Aviation AdministrationCalgary Airports (YYC Calgary International Airport); Assaia
Who oversees AI-driven operations is contested as deployments scale
London's mayor wants a robotaxi cap and more TfL powers because TfL can only consent to or refuse AVs under a national licensing scheme; members of Congress ask the FAA to pause SMART over testing and training while DOT points to specialists and controllers who keep decision authority.
Operating questionDoes the agency that answers for the street or airspace have the legal power, data access and human-review design to oversee the AI system before it scales?
Supporting evidenceMayor of London / Transport for LondonU.S. Department of Transportation / Federal Aviation Administration
Evidence ledger
Full record · every source keeps its link and limitations
“It checks aircraft safety and even gives directions”... AI makes inroads across Incheon International Airport
Incheon International Airport Corporation describes AI already in operation across airside safety (AI video analysis that detects hazards, identifies aircraft and obstacles, and alerts on low visibility and lightning), passenger congestion forecasting for departure halls, security and immigration published to airport displays and Naver/Kakao, and 31 robots deployed in May, some with generative-AI conversation in six languages. It plans an Aviation AI Innovation Hub with a 5-MW GPU farm and a 40-MW-class data center. The article itself states the airport has not shown how accurately hazards are detected, whether forecasts shorten waits, or how much robots help, and calls for published usage and performance figures. This is a new, broader source than the archived 28 October award item: it adds the congestion, robot and compute plans and the measurement critique.
Why it matters, evidence and limitations
- Why it matters
- Relevant to public airport authorities weighing AI for airside safety, passenger flow and digital service, and to any that plan on-site AI compute. U.S. transferability is limited by different ownership, airside regulation (FAA rather than Korean authorities) and data-sharing practice with private platforms.
- Evidence and measured results
- Reporter's article based on the airport corporation's statements on 1 November 2026. No accuracy, false-alarm, wait-time or robot-usage figures are published; the only quantities are counts (31 robots of three types, six languages) and planned capacity (5-MW GPU farm, 40-MW-class data center). The critique of missing metrics is the reporter's; the data-governance concerns are attributed to unnamed aviation industry observers.
- Limitations and uncertainty
- Machine-translated by the publisher with an accuracy disclaimer. Operator statements are not independently verified. No measured outcomes. The May 2026 robot deployment and 28 October award are reported, not separately confirmed in this inspection.
Sadiq Khan calls for cap on numbers of robotaxis on London streets
London's mayor called for a cap on robotaxi numbers and more powers for TfL as the UK government prepares to allow fully driverless vehicles under a national licensing scheme. Under current rules TfL can only consent to or refuse AV use in London, not manage or license robotaxi services. Khan raised unanswered questions on air quality, congestion and driver livelihoods. Freenow by Lyft is testing about 24 Baidu Apollo Go vehicles in Brent and hopes to launch driverless service as early as next year; Uber offers limited Wayve rides with a safety driver under TfL licence. The reporter notes many test trips still require human intervention, with roadworks a difficulty.
Why it matters, evidence and limitations
- Why it matters
- Relevant to U.S. state DOTs, cities and transit agencies facing the same split between state or national AV authority and local street management. The specific legal framework is British; the transferable point is that the authority managing streets may lack powers over AV fleet size, data and operations.
- Evidence and measured results
- News report with direct quotations from the mayor and a Freenow by Lyft executive. The only figures are about 24 test vehicles in Brent and the company's statement that two-thirds of black-cab drivers are over 50. No safety, congestion or intervention-rate data are reported.
- Limitations and uncertainty
- Political statements and operator claims; no measured outcomes. The Guardian's own test drive is anecdotal. UK legal structure differs from U.S. state AV law.
DOT AI program easing flight traffic, control tower headaches — but still Dems oppose
The FAA's SMART tool (Strategic Management of Airspace, Routes and Trajectories), developed with Air Space Intelligence and deployed at three Washington-area airports in September, has helped reroute 110 flights in its initial stage, averaging more than 10 minutes and 56 gallons of fuel saved per flight and affecting 14,644 passengers, according to DOT. It combines about 200 data streams. DOT says aviation specialists monitor the tool and relay information to controllers, who make the decisions. DOT plans nationwide expansion; Reps. Ocasio-Cortez and Beyer have called for a pause, citing testing and training concerns and arguing it is no substitute for rebuilding the controller workforce.
Why it matters, evidence and limitations
- Why it matters
- Federal, not SLED, but it directly affects operations at publicly owned airports (Reagan National and Dulles under the Metropolitan Washington Airports Authority; BWI under the Maryland Aviation Administration). Airport authorities should understand how upstream AI planning changes departure timing and gate demand.
- Evidence and measured results
- DOT/FAA-reported initial results relayed by the reporter: 110 rerouted flights, more than 10 minutes and 56 gallons saved per flight on average, 14,644 passengers, some flights up to 30 minutes. No baseline method, counterfactual or independent evaluation is described. The article is openly favorable to DOT, quotes an unnamed source, and summarizes the critics' letter. The FAA's SMART one-pager (opened as corroboration) describes SMART as a cloud-based platform using AI and says initial operations begin in fall 2026.
- Limitations and uncertainty
- Results are early, self-reported by DOT, and relayed by a partisan-leaning article; savings methods are not explained. 110 flights is a small sample. The 'source told The Post' commentary is unattributed. Federal ATC is outside SLED operations.
TEPA Seeks Over Rs590 Million for AI-Based Traffic Signals Across Lahore
Lahore's Traffic Engineering and Planning Agency has submitted a formal summary asking Punjab's Housing Department for more than Rs590 million to install AI-based traffic signals with vehicle video detection across the city. Cameras at intersections would assess traffic pressure in real time and adjust signal timings automatically to volume and flow. The request follows what officials call the successful completion of a pilot at Adda Plot and Jati Umra Chowk, where modern pelican signals were activated. Major intersections would be integrated gradually.
Why it matters, evidence and limitations
- Why it matters
- A city traffic agency moving from pilot to a funded citywide adaptive signal program is a common U.S. pattern. Transferability is limited: Pakistan's funding and approval channels and signal standards differ, and no performance data are provided.
- Evidence and measured results
- Wire report citing 'official sources' and the agency's funding summary. The pilot is described as successful but no delay, travel-time, safety or reliability results are reported. The pilot sites used pelican (pedestrian) signals, and the article does not say whether the pilot tested the AI detection system itself.
- Limitations and uncertainty
- Short wire report; unnamed sources; no measured outcomes; the relationship between the pelican-signal pilot and AI detection is unclear. Funding approval has not been reported.
TransCore Launches Open-Road Tolling at DRJTBC's New Hope Bridge
TransCore says it has launched open-road all-electronic tolling at the New Hope-Lambertville Toll Bridge (U.S. Route 202), the first of DRJTBC's seven legacy toll bridges to convert, replacing cashless tolling through existing plaza infrastructure. The installation includes TransCore's Infinity Digital Lane System and VCATS, its AI-powered vehicle classification solution. The bridge carries slightly more than 10,000 vehicles a day with nearly 94% E-ZPass use, the highest in the system. TransCore says the change improves safety and reduces long-term maintenance; DRJTBC's executive director calls it a milestone and the setup a framework for the remaining six facilities.
Why it matters, evidence and limitations
- Why it matters
- Directly relevant to toll authorities converting to open-road tolling, where AI vehicle classification replaces plaza-based classification. A bi-state U.S. authority with a planned six-facility migration makes this a reference for comparable conversions.
- Evidence and measured results
- Vendor announcement republished by a trade site, with an executive-director quote. Traffic volume and transponder share are given; no classification accuracy, revenue leakage, image-review rates or safety data are reported. Safety and maintenance benefits are vendor claims.
- Limitations and uncertainty
- Vendor-authored claims republished on a third-party site; no performance figures for AI classification; no independent evaluation. Original TransCore or DRJTBC releases were not separately inspected.
YYC first airport in Canada to deploy ApronAI in cargo operations
Calgary Airports is extending Assaia's ApronAI, already rolled out at YYC passenger gates last year, to cargo operations. AI-enabled cameras are installed at cargo Apron 9 and go live on 10 November 2026. The system will track each freighter turnaround and log when unloading, loading and fuelling begin and end, with Calgary Airports, cargo operators and ground handlers seeing the same real-time information. The project received federal National Trade Corridors Fund support. The airport's CIO says going live before peak season is intentional and lessons will inform 2027 planning.
Why it matters, evidence and limitations
- Why it matters
- Relevant to public airport authorities seeking shared operational data with tenants and handlers, and to agencies using federal freight-corridor funding for AI. Canada's airport-authority model is similar to some U.S. airport authorities, but U.S. grant programs differ.
- Evidence and measured results
- Vendor release with quotes from the airport CIO, Canada's transport minister and Assaia's CEO. Contextual figures: more than 5,200 cargo landings last year, more than 70% of Alberta's air cargo tonnage, more than $100 million in cargo infrastructure. No results from the passenger-gate deployment or targets for cargo are reported.
- Limitations and uncertainty
- Vendor-authored; go-live is future (10 November 2026); no measured outcomes from the earlier gate deployment are given.
KAI Introduces AI Chatbot Kaila, Serves Customers 24 Hours
KAI introduced Kaila, an AI-based virtual assistant integrated with its KAI121 service, providing 24-hour train travel information including schedules, routes and classes, with reservations planned later. KAI's commercial director said Kaila also serves as a means to collect more comprehensive consumer behavior data to shape services, and that AI capabilities will be improved gradually.
Why it matters, evidence and limitations
- Why it matters
- Relevant to public transit and rail agencies considering AI customer-information assistants. The explicit aim of collecting consumer behavior data is a governance flag for U.S. agencies subject to public-records and privacy rules. Transferability is limited by Indonesian law and KAI's commercial model.
- Evidence and measured results
- Report of an executive's remarks on 29 October 2026. No usage, accuracy, containment or satisfaction metrics. The reservation function is future.
- Limitations and uncertainty
- Machine-translated by the publisher with an accuracy disclaimer; executive statements only; no metrics; limited detail on technology or vendor.
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.
- Vendor claim
- A supplier-provided assertion that has not been upgraded to independent evidence.