The problem
On 23 October NHTSA announced a proposal to rewrite parts of FMVSS No. 108, the federal lighting standard, for automated vehicles. The Federal Register published it on 26 October, and comments are due on 25 November. Most of it removes requirements that assume a human driver, such as manual turn-signal levers and beam switches on cars that have no driving controls. The part that got the headlines is new: an optional cyan “ADS marker lamp” that tells police and other road users an automated driving system is driving.
The coverage focused on the colour. My conclusion up front: the colour is the least important decision in the document. What matters is how the lamp is defined. NHTSA does not define it as “the ADS is switched on”. It defines it as a lamp that indicates the vehicle “is operating without the expectation that an occupant is monitoring the roadway”, and it requires the lamp to switch off automatically once that expectation returns. That makes the lamp a public statement about who is responsible right now. Any company that fits one has to say, in engineering terms, the exact moment that responsibility moves.
What I concluded
First, the definition draws a visible line between supervised and unsupervised products. A driver-assistance system that requires the human to watch the road can never light this lamp, however capable it is, because the expectation that someone is monitoring never goes away. A driverless robotaxi lights it whenever it drives. NHTSA’s text also says that, because the rule would explicitly permit these lamps under its activation criteria, it would prohibit fitting them to vehicles without ADS functionality. In practice, a cyan light on the street would mean “no one inside is expected to be watching”.
Second, the hard case is the conditional system in the middle. For a car that drives itself in a traffic jam and then asks the human to take over, the lamp has to go off at some point during that handover. The proposed text says only that the lamp “must be automatically deactivated when a vehicle is operating with the expectation that an occupant is monitoring the roadway”. It does not say whether that moment is when the takeover request starts, when the human confirms, or when the request times out. For a robotaxi the question doesn’t arise. For a conditional system it is the whole product spec, and it is exactly what a police officer walking up to the car needs to know.
Third, NHTSA chose standardisation over a mandate, and said why. The agency writes that it “lacks sufficient information to conclude that there is a safety need” to require the lamp on every ADS vehicle, and that it cannot estimate the safety benefits. Its stated aim is narrower: if companies are going to add lights anyway, they should all look the same and not drown out the required lamps. That is a reasonable first step, but it means the lamp’s value depends entirely on how many operators choose to fit it.
Constraints
The first constraint is the existing lights. NHTSA deliberately set the brightness lower than the SAE J3134 industry standard, because it tentatively concluded the SAE values would interfere with turn signals and other required lamps. A front lamp would need at least 4 cd with headlamps on and 20 cd with them off, with a 125 cd ceiling, which is the same maximum as a parking lamp. It must burn steadily, all lamps must switch on and off together, and it may not be optically combined with any other lamp, including daytime running lamps. There must be one or two lamps on each of the front, sides and rear, so four to eight in total, mounted at least 15 inches high.
The second constraint is that the lamp is not allowed to say anything else. Some designers would like a light that also signals intent, such as “I’m about to pull out”. NHTSA proposes steady burning only, and says it is “not aware of any research supporting this usage”. It asks for research-based comments rather than allowing it now.
The third constraint is federal versus state law. NHTSA notes that at least one state, California, already permits these lamps under SAE J3134 (Cal. Vehicle Code §§ 25950 and 38750). It tentatively concludes that its rule would preempt the California provisions wherever its specification differs from J3134, for example on brightness, lamp combinations and manual switching. It also asks whether states should be barred from requiring the lamp. So a fleet that already runs to the California spec may have to change hardware.
The fourth constraint is evidence. The demand case rests on surveys, not crash data. An international survey of more than 4,400 drivers in the UK, Japan, the US and Germany found majorities in every country in favour, upwards of 80 percent in the US and the UK. Mercedes-Benz ran a two-year on-road validation in California from 2023 to 2025 with five vehicles. Meanwhile the UNECE working party on automated vehicles recommends uniform provisions for such a signal but does not recommend requiring it, citing uncertainty and the need for more research.
The options, and what I traded away
My reconstruction of the choice set, not a description of anyone’s internal reasoning.
| Option | What it buys | What it costs | Why I would / would not pick it |
|---|---|---|---|
| Mandate the lamp on every ADS vehicle | One unambiguous signal on every driverless car | A mandate NHTSA itself says it cannot yet justify with safety data | Not yet |
| Optional, standardised “if equipped” lamp (what was proposed) | Consistent look, protects required lighting, legal certainty for those who fit it | Coverage depends on each operator’s choice | I would pick it, with a tighter switch-off rule |
| Leave it to SAE J3134 and the states | No federal work; California already allows it | A patchwork of colours and brightness across states | I would not |
| Allow flashing to signal intent | A richer message to pedestrians | No research supports it yet, and it mixes two meanings into one light | Not until there is research |
| Tie lamp state to a logged responsibility state | The lamp becomes auditable after a crash | Adds a data requirement to a lighting rule | I would add it, through guidance if not the standard |
Evidence
The strongest evidence is the regulatory text itself. The definition and the activation rows in the proposed tables both use the same wording about the “expectation that an occupant is monitoring the roadway”, and the rows say the lamp “may be automatically activated” in that state and “must be automatically deactivated” outside it. NHTSA also lists every place it departs from SAE J3134. One is notable: SAE says the occupant must not be able to switch the lamp on or off manually, and NHTSA omits that requirement and asks for comment. The industry reaction so far is supportive. The Verge quotes Waymo welcoming the effort to “remove requirements that don’t apply to vehicles designed without a human driver”, and Nuro calling it “a positive step”. The Verge also reports that Mercedes-Benz, which won state approval for turquoise lights on its Level 3 Drive Pilot, has since moved to a Level 2++ system. If that is right, the company that did the most to push the lamp may not sell a car that can use it.
Where this stops being true
Segment: for driverless robotaxis with no driving controls, the switch-off question is trivial and the lamp is mostly a branding and policing aid; my argument applies to conditional systems that hand control back to a person. Region: in the UNECE markets, where the working party does not recommend requiring the lamp, the US rule may become the reference design or may diverge from what Europe eventually adopts. Fleet size: with a few thousand driverless cars in a handful of cities, a voluntary lamp may matter little; if driverless cars become common on mixed roads, the pressure for a mandate will rise and the definition will carry more weight. Evidence: if research shows road users do not change behaviour when they see the lamp, the safety case for it weakens and it becomes a courtesy light. And if the final rule specifies the switch-off moment precisely, my main criticism goes away.
What I would do differently
Define the switch-off moment in the rule, not in each company’s design notes. The simplest version: the lamp goes off when the system issues a takeover request, not when the human finally responds. Road users should never see cyan on a car whose system has already asked a person to take over.
Keep SAE’s ban on manual switching. A lamp an occupant can turn on or off is a lamp nobody outside the car can trust, and it defeats the point of defining it by responsibility.
Log the lamp state next to the automation state. If a crash report shows the system engaged but the lamp off, or the lamp on after a takeover request, that mismatch should be visible to investigators.
Hold the line on intent signalling until there is research. One light meaning one thing is the whole value of a standard colour.