09How traffic lights work

The short version: A traffic light is a small computer in a roadside cabinet, a set of sensors that tell it who is waiting, and a separate safety box that will switch the whole intersection to flashing if anything looks wrong. The computer divides time into phases, using a diagram engineers call rings and barriers, so that conflicting movements are never green together. Many city signals run timing plans set from traffic counts and pushed out from a control room; some retune themselves continuously. Whether the crosswalk button does anything depends on where and when you press it.

How it changed: from a gas lamp in London to a computer on every corner

The first traffic signal stood outside the Houses of Parliament in London, near the House of Commons at Great George Street and Bridge Street.¹⁸² It was designed by John Peake Knight, a railway signaling engineer, and borrowed the railway's vocabulary: semaphore arms by day, red and green gas lamps by night, worked by hand by a police constable.¹⁸³ Sources disagree on the exact day it went up: Wikipedia gives 9 December 1868 and History Hit gives 10 December 1868.¹⁸⁴¹⁸³ It did not last.

A gas leak caused it to explode in 1869 - Wikipedia dates this to 2 January - and the policeman on duty was badly injured.¹⁸²¹⁸⁴ History Hit says the operator was "reportedly" killed, but the more careful sources say injured.¹⁸³ The signal was gone by 1870.¹⁸²

Electricity brought the idea back. In Salt Lake City, police officer Lester Wire developed an early electric traffic light in 1912.¹⁸⁴ On August 5, 1914, Cleveland's safety director Alfred Benesch tested a red and green signal from the American Traffic Signal Company at East 105th Street and Euclid Avenue. Lights about 15 feet (4.5 m) up on four corner poles showed "stop" and "move"; an officer in a booth threw the switches by hand, and bells rang to warn of a change, twice for Euclid and once for East 105th.¹⁸⁵

Yellow arrived in Detroit in 1920. William L. Potts (1883-1947), a Detroit police officer, designed the first three-color, four-direction signal, installed at Woodward Avenue and Fort Street.¹⁸⁶ Earlier lights had only red and green; three colors were standard by the mid-1930s.¹⁸⁶

Linking signals together came almost as early as the signals themselves. Salt Lake City tied six intersections to one manual switch in 1917; in 1922 Houston added automatic timers made by Crouse-Hinds; in 1952 Denver used one computer to control 120 lights; and in 1967 Toronto ran 159 signals from computers that took in vehicle detection.¹⁸⁴ Britain trialed automatic signals in Wolverhampton's Princes Square in 1927 and installed its first permanent ones at Park Row and Bond Street in Leeds in March 1928.¹⁸⁴

Did Garrett Morgan invent the traffic light?

No, though he did invent a traffic signal, and the distinction matters. Garrett Morgan, a Black inventor and businessman,¹⁸⁷ was granted US Patent 1,475,024 for a three-position traffic signal on November 20, 1923.¹⁸⁴ His design was a manually operated post with "stop" and "go" arms, cranked by an attendant. Its third position stopped traffic in all directions so the intersection could clear, and he sold the rights to General Electric for $40,000.¹⁸⁷¹⁸⁵ The Cleveland Police Museum notes that two-way stop-and-go signals already existed when Morgan designed his.¹⁸⁵

So the popular claim that Morgan invented "the traffic light" or "the yellow light" does not hold up: London had a signal in 1868, Salt Lake City and Cleveland had electric ones by 1912 and 1914, and Potts' three-color light dates from 1920. In our reading, the accurate version is still a good story. Morgan patented and commercially licensed a signal with an all-stop position, a feature that resembles the all-red clearance time modern signals still use, and he did it as a Black businessman in 1920s America. He deserves credit for that, not for an invention that predates him.

What's in the box on the corner?

A signalized intersection has three working parts you can see from the sidewalk:

• Signal heads: the housings on poles or mast arms that hold the red, yellow and green lights.
• The controller cabinet: the gray or green metal box at one corner, holding the electronics.
• Detectors and push buttons: the sensors that tell the cabinet who is waiting.

Inside the cabinet is the controller, the intersection's "brains": a small, ruggedized computer that decides which movements get green and for how long.⁷⁴ Next to it sit load switches, which turn the controller's 12-volt DC logic signals into the 120-volt AC power that lights the lamps, plus a flasher that produces flashing operation.⁵⁰

US controllers come from two family trees. NEMA TS 1 (1976) defined basic actuated control so equipment from different makers was interchangeable, and NEMA TS 2 (1992) added a serial data link and defined coordination and preemption. A parallel line ran from the hardware-only Type 170 to Caltrans' Model 2070 (1995), which also defined the processor, operating system and software language. The current Advanced Transportation Controller (ATC) family is meant to merge them in an open architecture.⁵⁰

The lamps changed around 2000. A paper from that time put LED signals at roughly 80-90% less energy than incandescent bulbs, lasting about 5-10 years instead of about one; about 400,000 LED modules were then installed in the US, out of roughly 11 million signal lamps.¹⁸⁸ We found no current count of how far LEDs have spread since. (Streetlights went through a similar switch; see How streetlights work.)

Why do traffic lights flash red?

The most important box in the cabinet is the one that does not run the intersection. The conflict monitor (in TS 1 cabinets) or malfunction management unit, or MMU (in TS 2), is separate hardware that watches the controller's outputs independently.⁷⁴⁵⁰ If it ever sees two conflicting movements green at once, or voltages that should not be there, it overrides the controller and puts every signal head into flash.⁷⁴

This is the fail-safe: the worst a computer fault should produce is a flashing intersection, never green both ways. In the US, agencies commonly set the flash either to red on every approach, which drivers treat as an all-way stop, or to yellow on the main road and red on the side streets.

Delaware's state edition of the MUTCD, for example, recommends flashing yellow for the major street and flashing red for the other approaches unless flashing red is used on all of them, and requires every green indication to go dark during flash.¹⁸⁹ State traffic law gives flashing red the force of a stop: Iowa's code says a flashing circular red means traffic shall stop, then may proceed cautiously while yielding to vehicles not required to stop, and a flashing yellow means proceed with caution.¹⁹⁰

Flash is also used on purpose. Many Boston signals flash overnight, when traffic is light.⁴⁷

Phases, rings and barriers, explained without the jargon

A signal divides time into phases, each a block of time serving a set of movements that can safely go together, such as northbound and southbound through traffic. One full trip through all the phases is the cycle length.¹⁹¹ Each phase's share of the cycle, including its yellow and all-red clearance time, is its split.¹⁹¹

US engineers draw this with the NEMA ring-and-barrier diagram. A ring is "a sequence of two or more conflicting movements" timed one after another; a typical controller runs two rings side by side, each with up to four phases.¹⁹² A barrier is a point where the two rings must line up, so that conflicting movements in different rings are never selected and timed together.¹⁹²

In plainer words: think of the left side of the barrier as "main street's turn" and the right side as "side street's turn". On the main-street side, ring 1 might serve the southbound left turn and then northbound through traffic, while ring 2 serves the northbound left and then southbound through traffic. Each ring can move at its own pace - a busy left-turn lane can get more time than the opposite one - but both rings have to arrive at the barrier together before the side street gets anything. That is how the controller stays flexible without ever giving a left-turning car a green into oncoming traffic.

There are two broad ways to run the clock:

• Pretimed (fixed-time) control serves every phase every cycle for a set time regardless of traffic. It is typical in closely spaced downtown grids.¹⁹²
• Actuated control uses detectors to decide whether a phase is served and how long it lasts. Fully actuated signals have detection on every approach.¹⁹² Many real intersections mix the two: a coordinated main street, with side streets and left arrows served only on demand.

What are the cut lines in the pavement at a red light?

Those rectangles, diamonds or circles of sealed saw cuts near the stop line are usually inductive loops. Wire is laid in slots sawed into the pavement and sealed over.¹⁹³ When a vehicle sits above the loop, its metal reduces the loop's inductance, which raises the frequency of an oscillator in the detector electronics; the detector card in the cabinet reads that change as a "call".¹⁹³

Agencies also use other sensors, each with trade-offs:¹⁹³

A loop's weak point is anything with little metal. A bicycle may not register on a loop that is not set sensitively enough, which is why people on bikes sometimes wait at a red that never changes, and why some cities paint a symbol showing where to stop a bike over the loop.

Do crosswalk buttons actually work?

Sometimes. In busy downtowns, many signals use pedestrian recall: the walk signal comes up automatically every cycle, so the button has no effect on timing. Boston uses recall in its busy pedestrian areas.⁴⁷ In 2017 the Boston Globe reported that "the vast majority" of buttons in Boston's downtown neighborhoods do nothing by design; some are inactive by day but work overnight, buttons outside downtown generally work, and the dead ones remain because removing them costs money.¹⁹⁴ The same report said New York City had removed several hundred nonfunctioning buttons in areas under reconstruction but kept others.¹⁹⁴

In 2020 Brookline, Massachusetts deactivated push buttons at 55 signals, and Cambridge and Arlington followed; Providence made its change permanent.¹⁹⁵ MassDOT kept the buttons working at its roughly 1,400 pedestrian signals, and Boston pointed out that many downtown signals already gave the walk automatically.¹⁹⁵

In a dense core during the day, the button often does nothing because the walk is already coming. In suburbs, at night, and at mid-block crossings, it usually does call the walk signal. On an accessible signal, pressing the button can still matter even where timing is automatic, because it can trigger the audible and vibrating cues described below. Countdown timers, which spread in the 1990s, are now required in the US on new or upgraded signals at wider crossings; US rules prohibit countdowns for drivers.¹⁸⁴

Leading pedestrian intervals and accessible signals

A leading pedestrian interval (LPI) is "a brief (3-7 second) advance WALK indication" before parallel traffic gets its green.¹⁹⁶ The head start puts people a few steps into the crosswalk, where turning drivers can see them, before cars move. FHWA studies cited by Caltrans show a 13% reduction in pedestrian-vehicle crashes at intersections that use them.¹⁹⁶

An accessible pedestrian signal (APS) is a crossing device for blind and low-vision people. Boston's specification shows what one does: a locator tone about 0.15 seconds long, repeating every second and adjusting its volume to street noise, so a person can find the button; a button that vibrates during the walk; an arrow pointing in the direction of the crossing; and an optional spoken message. The messages can be programmed over Bluetooth.¹⁹⁷

In American Council of the Blind v. City of New York, a federal court ordered New York City to install APS at 10,000 intersections within 10 years and at all signalized intersections by 2036.¹⁹⁸ At the time, about 749 intersections, about 5%, had them, out of more than 13,200 signalized intersections; the same advocates' page also gives 13,430.¹⁹⁸

How do green waves work?

A green wave, which engineers call progression, is a string of signals timed so that a vehicle at the planned speed meets green after green. Three numbers make it work.¹⁹¹

• The signals must share a cycle length, or a multiple of it (a quiet intersection can run two cycles for each one at a busy neighbor, which is called double-cycling).
• Each signal has an offset: the time between a master clock reference and the start of its own green, in seconds or as a percent of the cycle.
• The resulting bandwidth is the window of time in which vehicles can travel through the whole system at the set speed.

FHWA's 2008 timing manual prefers that cycles at conventional four-leg intersections "not exceed 120 seconds".¹⁹¹ Plans are usually built from a software model of the corridor.⁷⁴ A green wave is easiest on a one-way street. On a two-way street with uneven block spacing, the timing usually has to favor one direction, often inbound in the morning and outbound in the evening. Many city signals run a few time-of-day plans (morning, midday, evening, overnight) switched by the clock.

Adaptive systems: signals that retune themselves

Adaptive signal control goes a step further and recalculates splits, cycle lengths and offsets continuously from live detector data. The two best-known systems date from the 1970s:

• SCATS, developed in 1975 by the New South Wales Department of Main Roads in Sydney, was "one of the first adaptive control systems to be available commercially". Its operator says it now runs at more than 65,000 intersections in 218 cities in 33 countries.¹⁹⁹
• SCOOT (Split Cycle Offset Optimisation Technique) was developed by the UK's Transport Research Laboratory in 1979 for the Department of Transport. It uses loop data to adjust splits, cycle times and offsets by region, with about a 15% improvement over fixed timing.²⁰⁰ Transport for London is replacing SCOOT with a newer system called FUSION, targeting 500 sites by 2025/26 and 4,350 by 2030/31.²⁰⁰

Adaptive systems are only as good as their detectors and their maintenance. They tend to help most where traffic is variable and hard to predict - around stadiums, malls or highway ramps - and add less where a well-tuned time-of-day plan already fits the traffic.

Preemption and bus priority

Preemption interrupts the normal cycle to give an immediate green to an emergency vehicle (the dispatch side is in From fire alarm boxes to 911) or to clear traffic off a railroad crossing, while showing red to conflicting movements.⁷⁴ One early documented case: in Houston in 1991-92, emitters on emergency vehicles that triggered signals within a quarter mile (400 m) at 22 intersections cut average emergency travel time by 16% and 23% in two fire districts.²⁰¹ The cost is that a preempted signal falls out of step with its neighbors and needs time to rejoin the green wave.

Transit signal priority (TSP) is gentler. The controller holds a green a few seconds longer or cuts a red a few seconds shorter for an approaching bus, staying within its normal pattern. In Boston, the city and the MBTA started a pilot in July 2024 at three intersections on Brighton Avenue (at Harvard, Linden and Allston streets) for bus routes 57 and 66. Bus waiting time at red lights fell 21%, and the two agencies announced an expansion.⁹⁴ (More on the buses themselves in How the subway and the bus actually run.)

Right turn on red

In most of the US, a driver may turn right on red after stopping unless a sign says otherwise. That became national rule because of the 1970s energy crisis: the Energy Policy and Conservation Act of 1975 effectively required states to allow right turn on red (RTOR) to keep federal energy assistance, and a 1992 amendment extended the idea to left turns on red from a one-way street onto another one-way street.²⁰² New York City has long prohibited it more broadly.²⁰²

The safety cost appeared quickly. A 1982 study in the Journal of Safety Research found that after mid-1970s adoption, pedestrian crashes involving right-turning vehicles rose 48% in New York State, 107% in Wisconsin, 57% in Ohio and 82% in New Orleans, and bicycle crashes of that type rose 72-82%. Drivers look left for a gap in traffic and miss people approaching from the right. Those crashes were 1-3% of all pedestrian and bicycle crashes.²⁰³

The debate reopened in the 2020s. A December 2024 Mineta Transportation Institute study found that most drivers roll through rather than stop, and recommended that states let cities ban RTOR by default and allow it only at selected intersections.²⁰⁴ New York City mostly prohibits it, except in parts of Staten Island.²⁰⁴ The rule is less common elsewhere: in much of Europe, turning on a red signal is not allowed unless a specific sign or arrow permits it.

Roundabouts: the intersection without a light

Sometimes the best signal is no signal. In a modern roundabout, entering traffic yields to traffic already circulating, at low speed. US studies of intersections converted from signals or stop signs show injury crashes down 72-80% and all crashes down 35-47%; at high-speed rural stop-controlled sites, injury crashes fell 85%.²⁰⁵ The Insurance Institute for Highway Safety (IIHS) estimates that converting 10% of US signalized intersections would have prevented 289 fatal crashes in 2024.²⁰⁵ The first modern US roundabouts were built in Nevada in 1990.²⁰⁵

The gains are not automatic. Two-lane roundabouts show smaller reductions and sometimes increases at first; in Washington State, crashes at two-lane roundabouts fell about 9% per year as drivers got used to them. Drivers failed to yield to pedestrians two to three times more often at multilane roundabouts than at single-lane ones.²⁰⁵ The safety gain comes from low speed and from removing right-angle and head-on left-turn collisions.

An FHWA safety team handout counts 32 vehicle conflict points at a four-way signalized intersection and 8 at a roundabout, which it describes as 75% fewer.²⁰⁶

Red light cameras: what the evidence says and who disputes it

Red light cameras photograph vehicles entering an intersection after the light has turned red, for ticketing. According to IIHS, 1,119 people were killed in red-light-running crashes in the US in 2024, about half of them pedestrians, cyclists or people in other vehicles.²⁰⁷

The research IIHS summarizes points toward fewer serious crashes, with a trade-off:²⁰⁷

• Large US cities with cameras had a 21% lower fatal red-light-running crash rate and a 14% lower rate of all fatal crashes at signalized intersections (Hu and Cicchino, 2017). In 14 cities that turned cameras off between 2010 and 2014, fatal red-light-running crash rates were 30% higher than expected.
• A seven-city FHWA-sponsored study found right-angle crashes fell 25% while rear-end crashes rose 15%; it judged the net economic effect positive because rear-end crashes tend to be less severe.
• A Campbell Collaboration review of 38 studies (Cohn and others, 2020) estimated 20% fewer injury crashes and 29% fewer right-angle injury crashes; a 2025 update by Hu and Cicchino gave 19%.

The number of US communities with camera programs has fallen since 2012, to 351 in 2025.²⁰⁷ IIHS is funded by insurers and supports camera enforcement. Opponents argue that longer yellow times achieve similar safety gains and that some programs are run for revenue. We have not reviewed their studies here, so we report the argument without weighing it.

An earlier federal study gives a more mixed picture. A 2005 FHWA evaluation of 132 camera sites in seven US jurisdictions found right-angle crashes fell 24.6% while rear-end crashes rose 14.9%, and concluded the cameras gave "a modest aggregate crash-cost benefit" of about $39,000 to $50,000 per site per year.²⁰⁸

Inside the traffic management center

Many cities run their signals from a traffic management center (TMC), a control room with camera feeds where engineers watch traffic, adjust timing and push plans out to the field. Central systems talk to field equipment using NTCIP (National Transportation Communications for ITS Protocol), a family of standards developed since 1996 by the Institute of Transportation Engineers (ITE), AASHTO and NEMA with the US Department of Transportation. NTCIP 1202 defines how a central station controls and monitors actuated signal controllers and their detectors.²⁰⁹

When the link to the center fails, a signal does not go dark; it keeps running its own local timing. A Boston traffic engineer described what that means: "it's basically running free, out of sync with the rest of the signals in that corridor".⁴⁷

In Boston

The Boston Transportation Department (BTD) runs the city's signals from its Traffic Management Center on the seventh floor of City Hall.⁴⁷ In fiscal year 2016, BTD operated 854 signals, 557 of them (about 65%) connected to the center, with about 87% online on average. Engineers built timing plans in Synchro software.²¹⁰ By 2020 the city had "almost 900 traffic signals", 699 on the central computer system, more than 300 BTD cameras and access to about 500 more from police and MassDOT.⁴⁷ State agencies also run signals on their own roads inside the city, so BTD's count does not cover every signal in Boston.

GBH's 2020 look inside the room described two to three engineers watching during rush hours, four default time-of-day plans (morning, off-peak, evening and overnight), a detailed survey of each signal every five years, and many signals flashing overnight.⁴⁷ An adaptive test was then planned for 47 signals in the Seaport.⁴⁷

A January 2018 Boston 25 News report said the state was seeking about $5 million to test the system, which could cover as many as 39 intersections.²¹¹ The city's Smart Streets page, last updated in October 2025, listed the Seaport adaptive signals network, planned with MassDOT, as "in design," with no intersection count.²¹² We did not find a report that it has gone live.

Newer layers include the MBTA bus priority pilot on Brighton Avenue⁹⁴ and Google's Project Green Light, which had expanded to 114 Boston intersections by May 2025.²¹³

What's changing now (2026)

Three changes are reshaping US signals, all incremental.

A new rulebook. The MUTCD (Manual on Uniform Traffic Control Devices), the federal rulebook for signals, signs and markings, was reissued as the 11th edition with a final rule published December 19, 2023 and effective January 18, 2024. States had until January 18, 2026 to adopt it, and Revision 1 was published and took effect March 5, 2026.²¹⁴

Signals that talk to cars. USDOT's National V2X Deployment Plan, released in August 2024, set a short-term goal of vehicle-to-everything (V2X) radios at 25% of signalized intersections in the 75 largest metro areas by 2028, and nationwide deployment in the 5.895-5.925 GHz band by 2036, according to trade press coverage.⁹⁸ The key message is SPaT (signal phase and timing), which tells a connected vehicle which light is showing and when it will change. NTCIP 1202 version 4, which ITE listed as still in balloting when we checked in October 2026, adds support for passing that timing data from controllers to roadside radios, including new concepts such as an "Assured Green Period", along with stronger network security.²⁰⁹

AI-suggested timing. Google's Project Green Light, launched in 2023, was active in 18 cities on four continents by May 2025. It uses AI and Google Maps driving trends to recommend timing changes, which city engineers can apply "in as little as five minutes". Google reports early results of up to 30% fewer stops and up to 10% lower emissions at intersections, as global figures.²¹³ These are company-reported, best-case numbers; we found no independent evaluation. The system advises human engineers rather than running signals itself.

How traffic lights differ around the world

Red for stop and green for go are near universal; the details are not.

• Britain shows red and amber together before green. In the UK Highway Code, "RED AND AMBER also means 'Stop'": do not move off until green. Amber alone means stop at the line unless you are too close to stop safely.²¹⁵ US signals go straight from red to green.
• Japan uses a deliberately bluish green. Old Japanese used the word ao for both blue and green, and the go light is officially ao. The commonly reported account is that in 1973 the government decreed the bluest shade of green allowed, so the lights still read as "go" internationally.²¹⁶ Japan, like the US, has not signed the 1968 Vienna Convention on Road Signs and Signals.²¹⁶
• Berlin kept East Germany's walking man. The Ampelmännchen was designed by traffic psychologist Karl Peglau on a 1961 commission from the East German transport ministry and debuted in East Berlin in 1969. After reunification, officials planned to replace it; a "Committee to Save the Ampelmännchen" won the right to keep it in 1997, and it has been a registered trademark since then.²¹⁷
• Turning on red is far more common in the US than in much of Europe, as described above; New York City is the large US exception.²⁰²

Report an error in this section