Driverless Operation: Engineering, Labor, and Public Acceptance

When Honolulu’s Skyline opened in June 2023, it became the first new US heavy rail line in more than a decade to enter revenue service with no driver in the cab and no attendant onboard. Twenty miles of elevated guideway, 21 stations protected by full-height platform screen doors, Hitachi Rail STS Communications-Based Train Control (CBTC), and an Operations Control Center (OCC) staffed by a small crew watching dozens of cameras at once. Honolulu was not a global breakthrough — Paris Métro Line 14 has run unattended since 1998, Singapore’s North-South Line was converted in 2015, and Dubai opened its first GoA 4 line in 2009 — but it was a domestic one. The combination of engineering, labor agreement, and political will that allowed Skyline to launch is the same combination every other US agency contemplating driverless operation has to assemble. This post unpacks each of those three pieces.

Why driverless operation is not a technology decision

Driverless operation is governed by the IEC 62290 Grades of Automation framework, where GoA 3 means no driver in the cab but an attendant onboard, and GoA 4 means no staff onboard at all. CBTC is a signaling and control technology that enables higher Grades of Automation, but CBTC and driverless operation are not the same thing. The vast majority of CBTC lines in the United States operate at GoA 2 — the driver remains in the cab, supervising an Automatic Train Operation system that handles routine acceleration, braking, and station stopping. Moving from GoA 2 to GoA 3 or GoA 4 is rarely a question of whether the signaling system can support it. The signaling system can almost always support it. The question is whether the agency, its labor partners, its insurer, its state safety oversight, and its political leadership can support it.

This post is for the program manager, the operations director, and the board member who needs to understand what a driverless project actually requires beyond the procurement of CBTC. It draws on the manuscript treatment of GoA 3 and GoA 4 in Chapter 8 of Communications-Based Train Control and on the documented operational record of the international and US deployments that have crossed the line.

The engineering prerequisites for GoA 4

A GoA 4 line is not a GoA 2 line with the driver removed. It is a system designed from the start (or comprehensively retrofitted) to absorb the loss of every safety function the driver had previously performed without compromise. The prerequisites are well established and uniform across vendors.

Platform screen doors are the most visible prerequisite. Every mainstream urban GoA 4 metro globally — Paris, Copenhagen, Singapore, Dubai, Barcelona, Milan, Honolulu — uses full-height or half-height platform screen doors. The reason is simple: with no attendant onboard and no driver to observe a falling passenger, the platform-to-track gap must be eliminated by engineering. Platform screen doors typically add $50,000 to $100,000 per door set per station to capital cost. On a 21-station line like Honolulu, that is meaningful but is not the largest line item.

Obstacle and intrusion detection sits at the next layer. Onboard forward-looking cameras and lidar detect track obstructions ahead. Trackside CCTV with video analytics watches platforms and tunnel sections for trespasser intrusion. Trackside intrusion detection on grade-separated guideway, fenced or otherwise, supplements onboard sensors. When any of these triggers, the response is automatic emergency brake, an alert to the OCC, and a hold until a maintenance crew has confirmed the line clear.

Onboard redundancy is more demanding than at GoA 2. Most US GoA 2 deployments specify 2-out-of-2 voting in the Vehicle On-Board Controller (VOBC); GoA 4 deployments commonly specify 2-out-of-3 voting so that a single processor failure does not force an in-revenue brake-to-stop. (For more on the redundancy architecture inside the onboard computer, see The Onboard Side of CBTC: Inside the VOBC.)

Two-way passenger emergency intercom in every car is non-negotiable. The OCC has to be able to talk to passengers immediately, and passengers have to be able to talk to the OCC. Without this, the OCC cannot triage a fire, a medical emergency, or a security incident. Modern systems integrate this with automated smoke detection so that the OCC is alerted before any passenger acts.

The OCC itself is the silent prerequisite that often consumes more engineering effort than the train or the wayside. A GoA 4 OCC has to provide live CCTV feeds for every train and platform, real-time train telemetry, remote door open and close control, remote speed override and emergency stop authority, integrated dispatch to fire and ambulance services, and enough operator stations to manage the full fleet at peak with comfortable cognitive loading. A well-staffed OCC typically supervises 200 to 400 trains across multiple lines.

[FIGURE: Stack diagram of GoA 4 prerequisites — platform screen doors, obstacle detection, 2oo3 VOBC, two-way intercom, OCC visibility, redundant communication, automated yard, all sitting on top of CBTC. Source: existing book figure fig_08_04_goa4_prerequisites.png. Caption: “Driverless operation rests on a stack of independent prerequisites. CBTC is the bottom layer, not the whole system.”]

The labor question is the binding constraint in the United States

Most US transit operators are unionized. The Transport Workers Union, the Amalgamated Transit Union, and their local affiliates negotiate pay, work rules, staffing levels, and seniority. A move from GoA 2 to GoA 4 directly reduces the operator headcount the agency carries. The political and contractual cost of that reduction is severe in any state where transit unions are politically organized — which is most of the states where a meaningful CBTC retrofit would happen.

The pattern that has worked is binary. Either the agency commits to GoA 2 and preserves the operator role in the cab, or the agency commits to GoA 4 on a new line where there is no incumbent operator workforce to displace. GoA 3, with no driver but an onboard attendant, is the middle option that nearly nobody picks: the engineering cost is close to GoA 4, the labor savings are far smaller because the attendant still has to be paid, and the safety case is paradoxically more complex than GoA 4 because the attendant’s role and reaction time become a load-bearing assumption that has to be validated.

Honolulu is illustrative. Skyline is a new-build line on an elevated guideway with no incumbent rail workforce; the Honolulu Authority for Rapid Transportation could specify GoA 4 from day one without a labor displacement to negotiate. JFK AirTrain (2003), Las Vegas Monorail (2004), and Miami Metromover (1986) followed the same pattern: new-build, segregated, no displacement.

Any meaningful US main-line metro retrofit to driverless operation has to reckon with three labor realities at once. First, the existing operator workforce will not voluntarily exit, and forced separation triggers strikes that cost more than the long-term labor savings. Second, retraining the current workforce into station service, customer service, security, and maintenance roles preserves headcount but only partly preserves payroll savings. Third, attrition-driven workforce reduction over a decade is mathematically possible — the average operator tenure on US transit systems is shorter than 10 years on most properties — but requires sustained political and contractual commitment across multiple administrations and labor cycles. None of this is unsolvable. It is, however, the binding constraint, and any business case that ignores it is incomplete. (For the broader Grades of Automation context, see GoA 0 to 4 Explained: The Five Grades of Train Automation.)

Public acceptance is more tractable than agencies expect

The public acceptance question is the question every general manager asks early and gets the wrong answer to. The wrong answer is “Americans will never accept a driverless train.” The right answer, supported by the operational record of every line that has crossed over, is that public acceptance is a function of reliability, transparency, and emergency response visibility — not of an inherent cultural objection to automation.

Pre-launch surveys at Honolulu showed roughly 72 percent favorable opinion of driverless operation; post-launch surveys at the one-year mark showed 84 percent favorable. Paris Métro Line 1, the world’s busiest driverless line at over 750 million annual passenger journeys, took roughly 18 months from cutover to register higher passenger satisfaction than its manually operated peers in the same network. Singapore’s North-South Line conversion (2015) and East-West Line conversion (2019) both showed the same pattern: skepticism for the first weeks, neutrality by month three, and active preference by year two.

Three factors drive the curve. The first is reliability. Driverless lines do not call out sick, do not vary in performance, and hit their headway targets within tighter tolerances than human-driven lines. After roughly two months of incident-free operation, riders begin to notice. The second is transparency. Lines that publish their emergency response protocols clearly, post intercom locations visibly, and run public information campaigns explaining what happens if something goes wrong record acceptance figures 10 to 15 percentage points above lines that assume the rider will figure it out. The third is the emergency intercom itself. Riders consistently report that knowing they can press a button and reach a live OCC operator within seconds is more reassuring than knowing there is an attendant onboard, because attendants on attended lines are often elsewhere on the train when the rider needs help.

The acceptance pattern has held across cultural contexts that look nothing like each other: Paris, Copenhagen, Singapore, Dubai, Honolulu. The argument that the United States is unique on this dimension is not supported by the data.

What the engineering, labor, and acceptance picture means together

For an agency considering a driverless deployment, the planning sequence is well-defined.

The procurement specification has to bake in the prerequisite stack from day one. Adding platform screen doors, 2oo3 VOBC, obstacle detection, and OCC capability to a CBTC retrofit halfway through the project is order-of-magnitude more expensive than specifying them at RFP. (See Operating Modes in CBTC: Normal, Restricted, Manual, Failure for how these prerequisites cascade into the day-to-day operating mode envelope.)

The labor framework has to be negotiated before the procurement, not after. Agencies that wait until commissioning to engage their unions discover that the negotiation timeline can extend the project by 18 to 36 months and consume the savings the driverless decision was supposed to deliver. The agencies that succeed start with a multi-year transition agreement (retraining, attrition, no involuntary separations, station service expansion) and use the procurement to fund it.

The public communication plan has to start before the system is visible to riders. The Honolulu pattern — pre-launch awareness campaign, post-launch transparency on incidents and response, sustained executive presence — is the playbook. Agencies that treat public acceptance as a launch-day question rather than a multi-year cultivation report higher post-launch friction and longer ridership ramp-up.

Grade of Automation	Staffing	Capital Cost Relative Index	Labor Cost Index	Safety Case Complexity
GoA 2	Driver in cab	—	High	—
GoA 3	Attendant onboard	Comparable to GoA 4	Partial savings	More complex than GoA 4
GoA 4	No staff onboard	—	—	—

What this means in practice

• Driverless operation is an organizational decision wearing a technical costume. The CBTC system can almost always support GoA 4; the agency, its unions, its insurer, and its political leadership are what gate the decision.
• The engineering prerequisite stack — platform screen doors, obstacle detection, 2oo3 onboard redundancy, two-way intercom, OCC capability — is well-defined and uniform across vendors. Specify it at RFP, not later.
• GoA 3 (driverless with onboard attendant) is rarely the right answer. The labor savings are partial, the safety case is more complex than GoA 4, and the engineering cost is comparable. Most agencies that pick GoA 3 picked it because labor negotiations forced the compromise.
• US GoA 4 has worked on new-build segregated lines (Honolulu Skyline, JFK AirTrain, Las Vegas Monorail, Miami Metromover). Brownfield retrofit to GoA 4 on an established main-line metro is a multi-decade labor-and-engineering program, not a procurement.
• Public acceptance follows reliability, transparency, and visible emergency response. The international record across Paris, Copenhagen, Singapore, Dubai, and Honolulu is consistent: skepticism gives way to preference within roughly two years of incident-free operation.

Where to go next

This post is an 11-minute summary. The full treatment lives in Chapter 8 of Communications-Based Train Control (Volume 2): Buy on Amazon · Download Chapter 8 slides (free PDF)

Sources

• Wang, C. (2026). Communications-Based Train Control, Volume 2: Operations, Lifecycle, and US Deployment. Independent. ISBN 979-8-258-54295-3. — Chapter 8, “Grades of Automation.”
• International Electrotechnical Commission. IEC 62290-1: Railway applications — Urban guided transport management and command/control systems — Part 1: System principles and fundamental concepts.
• Honolulu Authority for Rapid Transportation. Skyline Project Overview and Operations Reports. honolulutransit.org
• RATP. Métro Line 1 and Line 14 Automation History. ratp.fr
• SMRT Corporation Singapore. North-South and East-West Line Renewal Programme — Driverless Operation Implementation. smrt.com.sg
• Federal Transit Administration. State Safety Oversight Program Standard. transit.dot.gov
• American Public Transportation Association. Standards for Communications-Based Train Control (CBTC) Systems.