Reference · Levels of Autonomy
"Fully autonomous" is the least useful phrase in this field. Autonomy is not a single number that a system either reaches or does not — it is a description of what a system can handle, in what conditions, with how much help. Getting that description right is the difference between a specification and a slogan.
The Problem With Levels
Ask how autonomous a system is and you will usually get a level between one and ten. The trouble is that a platform which can fly a pre-planned route over open ground in clear weather with no opposition, and a platform which can pursue an objective through cluttered terrain with its satellite navigation jammed, might be assigned the same number — because the number is answering different questions in each case.
A useful answer has to say autonomous at what, and under what conditions. That is precisely the gap the NIST ALFUS framework — Autonomy Levels for Unmanned Systems — was created to close, by giving requirements and evaluation a shared vocabulary instead of a marketing adjective.
The Three Axes
ALFUS describes autonomy as a point in a three-axis space rather than a rung on a ladder. Move along any axis and the difficulty changes, even if the platform does not.
How hard the task itself is. Holding a fixed orbit is not the same problem as searching an area, adapting to what is found, and re-planning against a changing objective. The same aircraft can be highly autonomous at one and useless at the other.
What the world is doing to the system. Open sky in daylight with a strong satellite fix is the easy end. Cluttered terrain, degraded weather, jammed or spoofed navigation and an intermittent datalink is the other, and it is where real operations live.
How much operator involvement is required, and of what kind. Continuous manual control sits at one extreme; a person setting an objective and boundaries and then supervising sits near the other. This axis is a design choice, not only a capability.
A claim on any single axis is close to meaningless without the other two. High human independence in a trivially easy environment on a trivially simple mission is not an achievement — and it is exactly what a demonstration video usually shows.
How We Specify It
We describe our systems along these axes rather than claiming a level, because a level invites the reader to assume the other two axes were hard. Ours are still in research and development, and we would rather be precise than impressive.
Our Position
We do not build toward maximum human independence. We build toward the point where autonomy removes workload without removing judgement.
Autonomy should absorb the parts of a job that consume attention without needing judgement — holding a pattern, watching a feed — and hand the rest to a person.
Freedom inside limits a person has set, and a stop at the edge of them rather than an improvisation past it.
Recognising that a situation has left the envelope the system should handle alone is itself a capability, and we treat it as a requirement.
No system we build selects and engages on its own authority, at any point on any axis.
The ALFUS framework is published by the US National Institute of Standards and Technology; this page summarises its structure to explain how we describe our own systems and does not imply any NIST endorsement, assessment or certification. All Rewind Dynamics systems remain under active research and development.
Go Deeper
If you are writing requirements for an autonomous system, we would genuinely enjoy the conversation.
