Almost every dispute about self-driving cars comes from people using one phrase for six different things. There is a standard scale, it is widely used by regulators and manufacturers, and knowing it makes marketing claims much easier to read.
The six levels, in plain terms
- Level 0. No automation. Warnings and momentary interventions such as emergency braking still count as level 0, because the driver is driving.
- Level 1. One thing at a time. Cruise control that manages speed, or lane keeping that manages steering, but not both together.
- Level 2. Steering and speed together, with the driver responsible at all times and required to watch the road. Nearly every system currently sold to consumers is here, whatever it is called.
- Level 3. The car drives in defined conditions and the driver may stop watching, but must take over when asked, within a reasonable window.
- Level 4. Fully driverless within a defined area or set of conditions. If a human does not respond, the car reaches a safe stop by itself.
- Level 5. Anywhere a competent human could drive, in any conditions. Nothing on the market is close.
Where the meaningful line sits
The important boundary is between level 2 and level 3, because it is where legal responsibility moves. At level 2 the driver is driving, even with hands off the wheel and the car steering itself. At level 3 the system is driving within its stated conditions.
This is why manufacturers are cautious about claiming level 3, and why names suggesting full autonomy usually describe level 2 systems in the small print. If the manual says you must supervise and be ready to intervene immediately, it is level 2 regardless of the branding.
Operational design domain, the term that explains the disagreements
Every system has a defined set of conditions in which it is designed to work: certain roads, speeds, weather, lighting and map coverage. A level 4 vehicle operating flawlessly in one mapped city is not therefore capable elsewhere, and a demonstration video shows performance inside a domain the viewer cannot see the edges of.
Most public argument about self-driving progress is really an argument about how wide these domains are, which is rarely stated in the marketing.
How to evaluate a claim
Ask three questions: what level is claimed officially rather than in advertising; what conditions does the manufacturer state the system requires; and who is responsible if it crashes while engaged. The third question is the most revealing, because responsibility follows capability in every regulatory framework that exists.
Common questions
Is a car that steers and brakes itself self-driving? Not in the sense that matters. If you must supervise, you are driving, and that is level 2.
Why are some driverless services already running? Because they are level 4 inside tightly defined mapped areas with extensive support infrastructure, which is a different problem from a car that drives anywhere.
Does more data automatically produce autonomy? Data helps with common situations. The difficulty is rare events, which by definition appear infrequently in data however much is collected.
What does the driver-monitoring camera do? It checks you are watching the road, which is a requirement of level 2 systems and an admission of what level they are.
Join the discussion