regolith
: material shaped by constant impact; fragmented, recombined, and refined over time into something stronger than its origins.
Regolith is the memory layer for autonomous physical systems. Today's software executes instructions; tomorrow's will remember the past to predict the future.
Modern industrial systems generate vast telemetry, yet almost every controller, robot, and automation system treats each decision as an isolated event — reacting to the current reading, rediscovering known failure modes, and losing hard-won knowledge every time a machine shuts down.
We believe persistent memory is the missing abstraction for autonomous industry, just as perception was for autonomous vehicles. Every physical system has a history, and that history should become part of how it thinks. We ingest sensor data, logs, and control actions into a persistent operational memory, then build state estimation, sensor trust, failure prediction, and autonomous control on top of it — so the question shifts from what is happening right now to what a system should do next, given everything it has experienced.
The Moon is not our first market. It is our hardest test case — vacuum, radiation, abrasive dust, and no easy access to repair. If software can run autonomous industry there, it can run almost anywhere: factories, robots, and the physical infrastructure the rest of the world depends on.
History
Regolith began by aiming at the Moon. We chose it deliberately: of all the places autonomous industry might one day run, it is the least forgiving — unmanned, hard to reach, and impossible to repair by hand. We reasoned that software able to operate reliably there could operate almost anywhere.
Working the problem in that extreme made the missing piece clear. What autonomous systems lacked was not sharper perception or faster control, but memory — a way to accumulate experience over a lifetime and let it shape what happens next. The essays and simulator here are artifacts of that early, Moon-facing work.
We've since pivoted to Earth as our initial case study. The same failures — degrading sensors, repeated faults, knowledge lost at every shutdown — already play out in factories, robots, and industrial equipment, with far more data to learn from. Earth is where we begin; the Moon remains the hardest test case we build toward.
Essays
- Metallurgy for a Vacuum World
How do we transform oxides into useful materials?
- Energy for a Vacuum World
How do we power industrial civilization?
- Memory for Autonomous Systems
How do systems remember?
- Trust for Autonomous Systems
How do systems know what is true?
- Control for Autonomous Systems
How do systems act?
Hiring
We are interested in people who think deeply about data, memory, control, and autonomy.
If our essays resonate with you, reach out: katiechai21 [at] icloud [dot] com