Driver monitoring (DMS)

A driver monitoring system watches the driver rather than the road. It infers whether the person behind the wheel is drowsy, distracted or looking away, and warns them when the answer says they should not be relied on to intervene. Early systems inferred all of this indirectly, from steering behavior and how long a hand stayed on the wheel; current ones use an infrared camera pointed at the face, tracking eyelid closure, head pose and gaze direction.

The distinction between those two generations is not a detail of implementation. It is the line the regulations are drawn along, and, in at least one case, the line a recall was written along.

Two regulated functions, often confused

European law does not regulate driver monitoring as a single thing. Regulation (EU) 2019/2144, the General Safety Regulation, requires two separate functions with different scopes and different dates:

• Driver drowsiness and attention warning (DDAW) — required for all new car registrations in the European Union since July 7, 2024.
• Advanced driver distraction warning (ADDW) — required for all new car registrations since July 7, 2026.

Drowsiness and distraction are not the same failure. A drowsy driver is degrading over minutes; a distracted one has looked away for seconds. DDAW can be satisfied without a camera at all, by inference from driving behavior. ADDW is the requirement that pushes an interior camera into the cabin of an ordinary car, because seconds-scale gaze direction is not reliably inferable from the steering column.

The two are routinely reported as one rule, and the difference matters commercially as well as technically: the supplier revenue that arrived in 2026 is attributable to the second date, not the first.

What the requirement is worth to the companies that supply it

Regulation of this kind shows up in a supplier’s accounts before it shows up in a standard. Smart Eye, a Swedish supplier of driver-monitoring software, reported second-quarter 2026 automotive revenue of SEK 93 million against SEK 42 million a year earlier — 128 percent organic growth — with license revenue, earned per vehicle built, up more than 200 percent in the quarter. Its chief executive attributed the royalty growth to the ADDW requirement. The company reported software in series production at 20 of its 24 automotive customers, against 15 a quarter earlier.

That is the clearest available measure of what a safety mandate is worth to the software behind it, and it is a more concrete fact about the regulation’s effect than the regulation’s own text.

Nobody has agreed how to test whether it works

The requirement to fit driver monitoring has run ahead of any agreed method for measuring whether a given system does its job.

In August 2026 the US National Highway Traffic Safety Administration stated plainly, in a Federal Register notice seeking clearance for a research study, that there is no consensus approach to how driver monitoring systems should be tested for drowsiness detection. It criticized the instrument the industry has been using: the Euro NCAP Safety Assist protocol specifies eyelid-closure, microsleep and sleep criteria, but not the driving task, driver characteristics or environment under which those criteria are to be measured. The notice added that such criteria capture late-stage drowsiness, where countermeasures may no longer be effective.

The study NHTSA proposed is small and physical — participants driving at 11pm, 1:30am and 4am on a test track near Phoenix after at least sixteen hours awake, and a simulator cohort in Iowa City answering sleepiness questionnaires every ten minutes. NHTSA said the work would not produce an objective test procedure for use in a Federal Motor Vehicle Safety Standard.

The American timetable is long. NHTSA’s revised New Car Assessment Program roadmap places driver monitoring for drowsy driving at 2035 for implementation, with research running 2027 to 2030, and lists research into driver monitoring for distraction as beginning in 2026. Europe has required the function since 2024 and 2026 respectively; the United States expects to finish deciding how to rate it in the middle of the next decade.

Enforcement through recall, not only through approval

China has demonstrated a third route, one that reaches vehicles already sold rather than vehicles seeking approval.

In August 2026 the State Administration for Market Regulation published a recall plan covering 2,740,642 domestically built Tesla Model 3 and Model Y cars. The regulator’s description of the defect was a description of a monitoring system’s limits: with assisted steering engaged, the existing attention-monitoring mechanism — steering-wheel torque — was not sufficient to prompt a driver whose gaze had left the road. The remedy was software alone, an over-the-air update adding in-cabin camera monitoring on top of the torque sensing already fitted.

This is the generational line drawn as an enforcement action: torque-based inference judged inadequate, camera-based gaze monitoring supplied to an installed fleet by software. It is also a demonstration that in a software-defined vehicle a monitoring requirement can be imposed retroactively, on cars built as far back as 2019, without a workshop visit.

How it sits beside the approval regulations

Driver monitoring is not a standalone regulation but a requirement that appears inside several.

UN R171, which approves driver control assistance systems, is substantially a regulation about the driver rather than the vehicle: it requires warning strategies that take effect when driver disengagement is detected, and its 01 series introduced an explicit vocabulary for re-engagement — a Hands On Request, an Eyes On Request, and a Direct Control Alert. A system that cannot tell whether the driver is engaged cannot implement any of them, which makes monitoring a precondition of the assistance function rather than an accessory to it.

In the European Union both the General Safety Regulation requirements and R171 reach a vehicle through the type-approval framework. Because the function is delivered in software and can be changed after sale, the update pathway matters as much as the original approval: whether a change to a monitoring strategy can ship without a new approval is a question for R156’s software update management system, and the Tesla case shows a regulator treating an update as the entirety of a remedy.

What to watch

The unresolved question is measurement, not requirement. A function has been mandatory in Europe for two years, is rated in a consumer crash-test program, and is now being retrofitted to millions of cars by regulatory order — while the largest vehicle-safety regulator in the world says on the record that no agreed way to test it exists, and does not expect to rate it until 2035.

Two consequences follow, and both are worth watching. The first is that compliance and effectiveness are further apart here than in most of vehicle safety, and the gap is currently filled by a single consumer-rating protocol whose own limits a regulator has now criticized in writing. The second is privacy: the requirement puts a camera pointed at the driver’s face into every new car in a large market, and the rules governing what that camera may record, retain and transmit sit in data protection law rather than in the safety regulations that mandated the camera. Those two bodies of law were not written together.