SHV GroupsSHV GROUPS
PlannedDivision

SHV Space, Defense & Robotics

The environments where autonomy matters most are the ones where asking for instructions is not an option. This division does not exist yet — this is what it is for.

Nothing here has started.

No work is underway in this division. There is no team, no programme, and nothing to request access to. What follows is why we intend to build it, written now so it can be held against us later.

A probe at Jupiter is forty minutes from an answer. It has to decide alone.

Space is an autonomy problem before it is a rocket problem.

Every kilometre of distance makes remote control less viable. Beyond a certain range a craft is not being operated at all — it is being trusted, and everything it does between transmissions it does on its own judgement.

That is the same problem we are already working on: reasoning that holds across a long horizon, adapts to an environment nobody could fully specify in advance, and does not drift from what it was sent to do. We are approaching it through benchmarks today because that is where it can be measured. The application is elsewhere.

Media slot — spacecraft or deep-space imagery

Where we think this leads.

Four directions, in rough order of how soon they could matter.

Eunoia Zeta

Earth observation

Satellites already record far more of this planet than anyone will ever look at. The bottleneck is not sensing, it is judgement — knowing which few frames out of millions someone needs to see today.

Eunoia Sigma

Climate systems

Climate is the long-horizon problem: slow, coupled, and punishing to model. Systems that reason about how a situation develops rather than how a sentence continues are the right shape of tool for it.

CRSTO

Robotics

A model that reasons well and cannot act changes nothing physical. Robotics is where a world model stops being a claim and starts being testable against the floor.

Deep-space autonomy

Missions that operate for years past the point of useful contact, making decisions their designers could not have enumerated. The furthest out, and the reason the rest is worth doing.

On the defense part of the name.

It sits in the division title, so it deserves a direct answer rather than a paragraph that avoids the question.

  • Any work in this area will operate under human control, with review that can stop it.
  • We will publish what we are prepared to defend, and decline work we are not.
  • Autonomy in high-stakes environments is a safety problem before it is a capability problem, and we intend to treat it in that order.
  • None of this is active. When it is, this page changes and says so.

Other divisions