Stelaris exists to build the layer the space economy is missing: the ability to physically act on things already in orbit.
If SpaceX is building the railroad, Stelaris is building the distribution layer — the FedEx of space.
As mass-to-orbit becomes affordable, the number of objects in orbit will grow by orders of magnitude. Today every payload carries its own propulsion, because every payload has to move itself. This is the equivalent of every shipping box having its own wheels and drive train. It does not scale.
The bet: in ten or fifteen years, constellations of autonomous robotic spacecraft — each with thrusters, grippers, and onboard coordination — will be dispatched to converge on objects of any mass, shape, or distribution, grab them, reposition them, and release them wherever they need to go. Hundreds or thousands of robots. Hundreds or thousands of simultaneous jobs. Heterogeneous, non-uniform, dynamically allocated. Shared infrastructure for an economy that is still being built.
A company built to serve that future has three pillars:
1. Algorithms — how the swarm decides, coordinates, and adapts. This is what we are building now. 2. Hardware — the physical robots, their actuation and sensing. 3. Logistics — how objects and robots are matched at the fleet level.
Pillar one is the one we can iterate on today, in high-fidelity simulation, with theoretical grounding from multi-agent reinforcement learning and swarm robotics. It is the hardest and most defensible layer, and the foundation the other two are built on. Without the coordination, the robots are just more debris.
Launch is nearly solved. Thrust is a commodity. The hard problem — the one that unlocks servicing, in-space assembly, debris removal, and logistics between orbits — is getting many autonomous units to agree, in real time and from only what each can sense locally, on how to grip, balance, and steer a shared mass none of them could move alone.
Whoever teaches a swarm to jointly manipulate the physical world owns the layer the entire space economy runs on. We are building that layer first.
We hold ourselves to a single standard: honesty about what is real.
A component earns credit only when an ablation proves it. We do not tweak something and guess whether it improved — we understand what changed, see it in the telemetry, and know why. What survives replication across environments, agent counts, and conditions becomes real. What does not is discarded or re-examined.
The research runs on an agent-driven process — a knowledge base, an experiment dispatcher, a daily research digest, and diagnostic pipelines — designed so the science scales without a human at every step. The same discipline that makes the research trustworthy is the discipline that will make the fleet trustworthy.