Where it breaks
A camera sees almost nothing useful in space without help. Sunlight arrives unfiltered and brutally hard, shadows fall to absolute black with no atmosphere to soften them, and the surfaces in frame are bare metal and foil that glare or disappear depending on the angle. The target is usually uncooperative: unlit on one side, carrying no markers, and slowly tumbling, so its appearance never holds still. The perception systems that handle this kind of scene on Earth were trained on a world that looks nothing like orbit, and they were trained that way because real labelled images of spacecraft barely exist. So they learn from synthetic renders and then meet the real thing, where they fail quietly, in ways that are hard to see coming and harder to trust. That distance, between what a system does on a bench and what it does in flight, is the entire problem, and it does not close with a bigger model. It closes through slow, measurable work: making perception survive real light, real motion, real sensors, and real consequences for being wrong.
The discipline
Stagral exists for that problem. Perception and autonomy for machines that have to operate in space without us, where a bad read is not a dropped frame but a lost vehicle, and where the system has to know the difference between seeing something and only thinking it does. The method is narrow on purpose: take one well-defined piece, build it properly, and test it against the conditions that actually break it rather than the ones that are convenient to simulate. All of it is done in the open and reported honestly, including the parts that fail, because in this kind of work the failures carry most of the information.
Posture
It is early and deliberate. There is no team and no roadmap, and the specific work is described elsewhere, as it is done. What is above is the position. The rest is being built.