Acting under uncertainty
The world model is never complete or perfectly correct. Good autonomy behaves sensibly anyway, and knows how confident it should be.
Capability 01 · Autonomy
Autonomy.
Autonomy is the part that decides. Given a goal and a picture of the world, it works out what to do next — and keeps working it out as the situation changes. It is the difference between a remote-controlled machine and a system that genuinely reduces workload.
What It Means
Not one thing, but a spectrum.
Autonomy is often talked about as a switch — either a machine is autonomous or it isn't. In practice it is a spectrum, and the useful question is always autonomous at what? A platform might navigate a route entirely on its own while still requiring a person to approve anything consequential.
We design deliberately along that spectrum. The goal is not to remove people; it is to remove the parts of the job that consume attention without needing judgement — holding a heading, flying a search pattern, watching a feed for hours — so the person can spend their attention on the things that actually need a human.
Inside The Capability
From goal to action.
The autonomy layer sits above perception and below the mission. It takes in what the platform understands about the world and turns a stated objective into a sequence of concrete actions.
- Mission reasoningTurning a high-level objective — cover this area, follow this route, watch this point — into an executable plan.
- Planning & replanningChoosing a course of action, then revising it when the world turns out differently from the plan.
- Behaviour arbitrationDeciding what matters most right now when several objectives compete for the same platform.
- Bounded discretionOperating freely inside limits a person has set, and stopping at the edge of them.
Autonomy LayerDecisions
InputMission goal + world model
OutputActions the platform executes
RunsContinuously, many times a second
Bounded byOperator-set rules and limits
StageResearch & development
Hard Parts
What makes autonomy
difficult.
Anyone can make a machine follow waypoints on a clear day. Autonomy earns its name in the cases that are not clean.
Knowing when to stop
Recognising that a situation has moved outside what the system should handle alone, and handing back to a person cleanly.
Predictability
A system that is brilliant but unpredictable is unusable. Operators need to be able to anticipate what it will do.
Graceful degradation
Losing a sensor or a link should reduce capability, not produce nonsense.
Explaining itself
Recording enough about a decision that it can be reconstructed and understood afterwards.
Real-time budgets
Every decision competes for the same limited onboard compute, and the loop cannot be allowed to slow down.
Everything described here is under active research and development. Nothing has been fielded, and we do not publish specifications or performance figures. Our first end-to-end mission test is planned for December 2026. Technical detail is shared directly with partners, evaluators and the armed forces under appropriate terms — get in touch.
Go Deeper
Want the detail
behind the autonomy?
We share architecture and approach with partners and evaluators under appropriate terms.