The Six Rungs of Autonomy – and Why the Top One Sits Empty
Waymo’s robotaxis are threading through San Francisco and Phoenix without a human hand on the wheel, fielding passenger trips, and making headlines every few weeks for something new. The company’s vehicles are classified as Level 4 autonomous – the second-highest tier in a formal six-level framework established by SAE International, the standards body whose definitions the auto industry and regulators now treat as the common language of self-driving development. What sits above Level 4, and why no vehicle on earth has reached it, is a question that reveals exactly how far the industry still has to go – and how strange the destination actually looks.
The SAE framework runs from Level 0 to Level 5. It is not a marketing scale. Each level describes a precise division of responsibility between the human occupant and the machine doing the driving. Most people who own a car today are living somewhere between Level 0 and Level 2, often without knowing it. Understanding the difference matters now because automakers, tech companies, and regulators are all using these levels to make decisions about what’s legal, what’s insurable, and what’s safe to deploy on public roads.
The gap between Level 4 and Level 5 is, in practical terms, the gap between a machine that drives itself very well in defined conditions and a machine that drives itself everywhere, in any condition, with no human available to intervene – ever.
Levels 0 Through 3: The Spectrum Most Drivers Already Know
Level 0 is a car that does nothing automatically. The driver handles steering, braking, acceleration, and all decisions. A conventional vehicle from the 1990s is Level 0. Level 1 introduces a single automated function – cruise control is the clearest example. The vehicle holds a speed, but the human manages everything else. Level 2 stacks two functions together: the vehicle can steer and control speed simultaneously, as in Tesla’s Autopilot or GM’s Super Cruise. The driver, however, must remain alert and ready to take over at any moment. The car is assisting. The human is responsible.
Level 3 is where the division of responsibility starts shifting in a meaningful way. At this level, the vehicle can manage the driving task fully under specific conditions – highway driving in clear weather, for instance – and the human does not need to actively monitor the road. But the human must be available to respond if the system requests a handoff. Mercedes-Benz received approval to operate Level 3 systems in Nevada and California, making it one of the first manufacturers to put a legally recognized Level 3 vehicle on American public roads. The critical liability question at Level 3 is response time: if the car asks you to take over and you aren’t ready, who bears responsibility for what happens next?
Level 3 has been the industry’s most contentious category precisely because it creates a middle zone of ambiguous human responsibility. You’re allowed to not pay attention – until you suddenly must. Honda’s Legend sedan carried a Level 3 certification in Japan. Outside of a small number of controlled deployments, the level has struggled to find wide commercial footing, partly because regulators worry about what happens when a distracted occupant fails to respond to a takeover request in time.
Level 4: What Waymo Actually Is, and What It Isn’t
Level 4 eliminates the takeover request problem by removing the human from the operational equation entirely – within a defined boundary. A Level 4 vehicle can drive itself without any human intervention, and if it encounters something it cannot handle, it does not ask a person to take over. It stops safely on its own. The human occupant has no obligation to drive, monitor, or respond. Waymo’s vehicles operate this way inside what the industry calls an Operational Design Domain – a geofenced area with mapped roads, known conditions, and defined weather and speed parameters. Outside that domain, a Waymo vehicle does not operate.
That boundary is not a flaw in Waymo’s engineering. It is the definition of Level 4. The system is designed to be fully autonomous within a specific envelope and to refuse operation outside it. Waymo has mapped its operating zones in painstaking detail – every lane, every traffic signal, every known construction zone. The vehicles rely on high-definition maps, lidar, radar, and cameras working together to navigate within that mapped territory. Add an unmapped road, an unusual weather event outside the system’s training parameters, or a geography the company hasn’t instrumented, and the car won’t go. That’s not a failure. That’s the system behaving as designed.
Zoox, Cruise (before its operational suspension following a 2023 incident involving a pedestrian), Baidu’s Apollo Go in China, and several other programs operate or have operated at Level 4 in defined urban zones. The pattern across all of them is the same: high capability inside a boundary, hard stop at the edge of that boundary. Level 4 is not a compromise on the way to Level 5. For most researchers and engineers in the field, it is likely where commercial autonomous deployment will live for the foreseeable future.
Level 5: The Definition That Makes It Unreachable
Level 5 requires a vehicle to perform the complete driving task under all conditions, on all roads, in all weather, at any time, without any human occupant capable of intervening – because there may not be a human in the vehicle at all. No geofence. No operational domain. No mapped territory requirement. A Level 5 vehicle would need to handle a dirt road in rural Montana during a blizzard as competently as it handles a mapped boulevard in Phoenix on a clear afternoon. It would need to navigate road types, signage conventions, pedestrian behavior patterns, and infrastructure conditions across every environment a human driver might encounter anywhere on earth.
The technical barriers to Level 5 are substantial and layered. Sensor systems that perform well in controlled urban environments degrade in heavy rain, snow, and fog – conditions that a human driver manages through experience and judgment built over years. High-definition maps, which underpin most current Level 4 systems, cannot practically cover every road globally and would require continuous updating to remain accurate. Edge cases – the unpredictable, the rare, the genuinely novel – are exactly where current machine learning systems struggle most. A child running into the road from an unexpected angle, a hand-painted detour sign at a rural construction site, a police officer waving traffic against a signal: human drivers handle these through general intelligence. Current autonomous systems handle them through pattern recognition trained on prior examples, and novel situations by definition have no prior examples.
There is also a definitional issue that some researchers raise: the SAE Level 5 standard was written as an aspiration, a theoretical endpoint that describes full machine equivalence to human driving under every possible condition. Whether that endpoint is achievable with any foreseeable technology – or whether it requires a form of artificial general intelligence that does not yet exist – is an open question in the field. Several engineers and researchers involved in autonomous vehicle development have argued publicly that Level 5, as strictly defined, may be permanently out of reach, and that Level 4 deployment across progressively larger operational domains is the realistic trajectory for the industry.
What the Levels Mean for Buyers and Riders Today
For anyone riding in a Waymo vehicle or purchasing a car with advanced driver assistance features, the SAE levels are the most honest tool available for understanding what the technology actually does. A Level 2 system marketed under a name like “Autopilot” or “Full Self-Driving” is still a Level 2 system – the human remains responsible and must stay engaged. Tesla’s Full Self-Driving package, despite its name, operates at Level 2 under SAE definitions. The branding and the standard do not always agree, which is where consumer confusion tends to accumulate.
Regulators in California, Arizona, and Nevada have begun requiring companies to specify the SAE level of their deployments as a condition of operating permits. The National Highway Traffic Safety Administration references the SAE framework in its guidelines on automated driving systems. As more vehicles with automated features reach the market, the levels will likely become as standard a reference point as miles per gallon or crash-test ratings – a number that tells you, precisely, what you’re getting.
The distance between a Waymo ride in San Francisco and a car that could drive you door-to-door from that city to a remote cabin in the Sierra Nevada without any mapped roads, in a January storm, is not just an engineering gap. It’s the distance between Level 4 and Level 5 – and right now, nobody has driven it.
