We have already explained how ADAS works — the driver assistance features fitted to almost every new car today. What we haven’t explained is the framework the entire industry, and every regulator, uses to talk about what comes after ADAS: the SAE Levels of Autonomy, a six-step scale (0 to 5) published by SAE International that defines, precisely, who — human or machine — is actually driving at any given moment. Understanding it is the difference between reading a headline about a “self-driving car” and knowing exactly what that car is legally and technically allowed to do without you.
The line that actually matters: Level 2 vs Level 3
The six levels split cleanly into two groups, and the boundary between them is the most important distinction in the entire framework. Levels 0 to 2 are all, formally, driver support systems: the human is legally driving at all times, required to supervise the system continuously and ready to intervene instantly, no matter how capable the car’s hardware is. Level 0 is momentary intervention only (basic automatic emergency braking); Level 1 automates a single axis, either steering or speed, not both (standard adaptive cruise control); Level 2 automates both simultaneously (adaptive cruise control plus lane centring, the combination sold today as “highway assist” or similar) — but the driver remains fully responsible and fully liable throughout. Levels 3 to 5 flip that responsibility: the system, not the human, is the one legally driving within its defined conditions. That single change — who is liable if something goes wrong — is why crossing from Level 2 to Level 3 has taken the industry the better part of a decade longer than the jump from Level 1 to Level 2 ever did.
Level 3: eyes off, but only just
A Level 3 system permits the driver to genuinely stop supervising the road — read a message, watch a video — but only inside a narrowly defined Operational Design Domain (ODD), and only if the driver resumes control within a few seconds when the system asks. Mercedes-Benz’s Drive Pilot, legally approved for use in Germany and in parts of the United States, is the clearest production example: it operates only in heavy traffic, below roughly 60km/h, on specific pre-mapped motorway sections. That narrowness is not a limitation the engineers failed to overcome — it is the entire point of Level 3: the moment the car takes over legal liability for driving, every single condition under which it does so has to be validated and certified in advance, which makes a wide ODD extraordinarily expensive to achieve and defend in court.
Level 4: no human needed, inside the box
A Level 4 system removes the human from the loop entirely within its ODD: no driver needs to be ready to take over at all, because the car is engineered to reach a minimal risk condition — safely stopping itself — if anything goes wrong, rather than handing control back to a person who might not be paying attention. This is the level robotaxi services like Waymo actually operate at, and it is also the level a large share of current EU-funded research targets directly. MODI (2022-2026, €23.0M, coordinated by ITS Norway) is explicitly named “a leap towards SAE L4 automated driving features”: it is demonstrating Level 4 automated trucks and shuttles on a real logistics corridor running from Rotterdam to Oslo, deliberately working through the unglamorous barriers — cross-border legal approval, infrastructure readiness, operational procedures — that decide whether L4 stays a demo or becomes an actual service.
Coordinating that kind of large-scale, cross-border demonstration work across Europe is itself hard enough to need its own dedicated research: FAME (2022-2025, €5.68M, coordinated by ERTICO – ITS Europe) is building the shared methodologies and evaluation tools that let projects like MODI, and the dozens of others running in parallel across different countries, actually compare results and share lessons learned instead of each reinventing its own validation approach from zero.
Level 5: the level nobody has built yet
Level 5 removes the ODD restriction entirely: full automation, everywhere, in every condition a human could drive in, with no steering wheel required at all. No production vehicle, and no serious research project, currently claims to be building Level 5 — it remains, for now, a theoretical endpoint rather than an engineering target, because validating “every condition anywhere” is a fundamentally different, and vastly harder, problem than validating the specific, bounded ODDs that Level 3 and Level 4 systems actually operate in.
Why the marketing rarely matches the level
Most of the confusion around “self-driving” traces back to one fact: a system’s name is not a legal claim about its level, but its owner’s manual is. Several driver-assistance packages marketed with names implying full autonomy are, in strict SAE terms, Level 2 — capable, sometimes remarkably so, but still legally requiring the driver’s continuous attention and instant readiness to intervene. The SAE scale exists precisely to cut through that: it doesn’t ask what a system is called, or even how good its hardware is, only one question — in this exact situation, is the human or the machine the one actually, legally driving? Everything else about how autonomous driving works — how the car senses its surroundings, decides what to do, and acts on that decision — is worth its own explanation, coming next in this series.
Photo: © Car-Shooters