Insight
Observations and opinions on how physical intelligence actually gets deployed.
Raising Machines in the Deep: Why Trial and Error Is the Costliest Thing at Sea
A robot vacuum eventually learns the shape of a room — bump by bump, until the map holds. A robot arm at the bottom of the ocean rarely gets that chance. It fails once, expensively, and the sea rarely lets it try again. This piece asks why smarter machines haven't made deep-sea robots more capable in practice — And the answer may change what actually gets built next.
Read →The Capability–Deployment Paradox
Demos keep getting more impressive; getting them to work on a real site has not got any easier. The gap is not about bigger models. It is the near-total absence of scene data and physics models built for a domain — the missing layer Sunnta builds.
Read →