General embodied intelligence emerges with compressing human action at scale.




We compress human behavior into Universal Action Representations, enabling interaction, adaptation, and physical intuition to transfer across environments and embodiments, while allowing new behaviors to emerge through continuous interaction beyond training.

For All
Forms Of Our
Future
Biped & Quadruped
Wheeled
Digital
Made forShaped byLife

We envision a future of embodied agents: intelligent machines capable of open-ended interaction with the physical world.

By unifying navigation, locomotion, and manipulation, our goal is to enable embodied systems to accumulate increasingly rich interaction priors across space, objects, environments, and time — ultimately powering intelligence that can autonomously adapt, infer intent, and discover entirely new ways of engaging with and delivering value in the world.

Tech Changes.
Values Don’t.
More Human, Not Less

We believe embodied intelligence should feel adaptive, collaborative, intuitive, and grounded in the realities of humanity — coexisting naturally alongside us across the physical and digital world.

The most important technologies do not pull humans away from life. They create more space for us to experience it.

Open Progress

The future of embodied intelligence will not be shaped by machine learning alone, but must emerge through collaboration across science, engineering, design, art, culture, storytelling, and the broader human experience.

The frontier of AI should not be built in isolation from the world it is meant to participate in.

Restrained Capability

In the physical world, every case is an edge case.

As embodied intelligence becomes more capable, restraint becomes just as important as capability itself. Knowing when not to act, when to adapt, and when to defer is fundamental to trustworthy real-world intelligence.

The goal is not intelligence without limits, but intelligence capable of understanding them.

The People
Building for People.

The real world doesn’t fit into one field. That's why our team spans AI, robotics, consumer tech, and software—with backgrounds including DJI, ByteDance, Tencent, Huawei, Roblox, Oxbridge, Tsinghua, and ETH.

And we're always looking for like-minded people.

Roger Jiang portrait
Founded by Roger Jiang, formerly of OpenAI and a co-inventor of ChatGPT and InstructGPT.
Beyond The Brain
Unified systems enable generalization.

We’re building the full-stack required for general embodied intelligence, bringing together multimodal learning, action modeling, whole-body control, and robotics engineering.

We’re intent on building intelligence across forms —
and into life itself.

Join Us
Contact Us
© Light Origins 2026Website by Fiddle.Digital