Levels of autonomy

Levels of autonomy is a scale that describes how much of the driving or operating a machine does for itself, and how much a person still has to do. It is used for cars, trucks, drones, farm machines and mining vehicles. The most common scale has levels from 0 to 5.

Here is how the scale works in general. At the lowest level, a person does everything and the machine only warns. At the middle levels, the machine takes over a part of the work, such as steering or keeping a distance, but the person must watch and be ready to take over. At the highest levels, the machine does the whole job in all conditions, and nobody needs to be at the controls. A machine at a higher level needs more sensors, more software and more proof that it is safe, so each step up is much harder than the one before.

The common scale is the one written by SAE (the Society of Automotive Engineers). Level 0 has no automation. At level 1 the machine helps with one task, such as speed. At level 2 it steers and controls speed, but the person watches all the time. At level 3 it drives in set conditions and asks the person to take over when needed. At level 4 it drives in set conditions and never needs the person inside them. Level 5 is the same everywhere. Level 3 is the hardest to make safe, because a person who has not been watching is slow to take over. This is why some companies, such as robotaxi developers, skip level 3 and build level 4 systems for a limited area. A few carmakers do sell level 3 systems, approved for use on certain highways. Indian road rules assume that a person is driving, so most automation on Indian public roads stays at the lower levels.

Consider a car with cruise control that keeps a set distance from the vehicle ahead. The driver still steers and watches the road, so this is a low level. Take, for example, a mining truck that follows a planned route inside a closed mine with no driver in the cab. The route is fixed and the area is controlled, so it can reach a high level even though the technology is not very different. Imagine a delivery robot on a college campus that picks its own path around people. It decides each next move itself, within a small mapped area.

The levels are about who is responsible, not only about technology. In lower levels the person is responsible for the driving. In higher levels, while the system is switched on, the system does all the driving, so more of the responsibility moves to the maker. Each country's law decides who is legally liable after a crash. The software behind this is described under driver assistance and the self-driving stack, and the idea of a machine that acts alone also appears in physical AI.

How is it different from safety certification? Autonomy levels describe what the machine can do. Safety certification is the proof, shown to a regulator, that it can do it safely.