What is NEURIX
Humanoid robots are not programmable.
They are extendable.
The Problem
Every humanoid platform ships its own software stack — its own SDK, its own control plane, its own assumptions about bodies and sensors. Capabilities written for one robot rarely survive the move to another. Skills are rewritten. Again and again.
Robotics progress is bottlenecked by this. Not by hardware, not by models — by the fact that nothing is portable.
The Concept
NEURIX is the extension layer for humanoid robots. It sits between reusable skills and the machinery of a specific body. Skills are written once and remain portable. Embodiment adapters bind them to each robot’s SDK, sensors, and kinematics.
A skill doesn’t know whose legs it is running on. The adapter does.
What is a Skill
A NEURIX skill is a self-contained package of behavior — perception, planning, control policy, and safety constraints — that executes under the runtime rather than against a particular robot. Skills are versioned, signed, and installable, like software.
Robots will be different. Skills should be the same.
The Two Laws
Two invariants are not features of NEURIX. They are its foundation. Everything in the runtime is built to uphold them, and every component that violates them is wrong by definition.
The robot sees only through sensors. The simulator obeys the same rules as the real body. Every motion passes through the same path — command, executor, live loop, policy. No hidden doors.