Physical AI, from ordinary video to real robots.

JOYX builds spatial intelligence — metric, three‑dimensional understanding of the physical world — from everyday video, and carries it all the way onto robots that work. The same engine runs live in broadcast sports today.

Working with Hitch Interactive, Overnght, USRowing and USA Water Polo. Built on AWS.

The real-to-real loop: it begins in the physical world — ordinary video becomes metric 3D, is fitted to a robot, verified against physics — and ends back in the physical world, on a real robot. Physical world the loop begins and ends here Ordinary video Metric 3D Fitted to the robot Verified against physics Real robot Real-to-real begins and ends in the physical world

Physical AI is gated by one thing: knowledge of the physical world that a machine can actually act on.

Video of the real world is everywhere and costs almost nothing. Understanding that is verified and true to scale is rare — and it is what a robot actually needs.

That is what JOYX builds. We turn ordinary video into spatial intelligence — a precise, three‑dimensional model of a scene and everything moving in it — and we carry that understanding through to real robots, in real environments, where it has to hold up.

Spatial intelligence is the foundation. Physical AI is where it goes to work.

Closing the real‑to‑real loop.

Most robot learning today depends on one of two sources, and each falls short on its own.

Teleoperation

Accurate, because a person is in the loop — but it cannot scale to the diversity of the real world.

Simulation

Scales without limit — but leaves a persistent gap between simulated and physical behavior.

Real‑to‑real

Takes a different path. Real‑world video and sensor capture is reconstructed into metric 3D, fitted to a specific robot, verified against physical constraints, and returned to a real robot as training data. The loop begins and ends in the physical world.

  1. Capture

    Ordinary video and sensor recordings of people and robots doing real work.

  2. Reconstruct

    The scene becomes a metric 3D model — geometry, motion and surfaces, at true scale.

  3. Fit to the robot

    Understanding is expressed for a specific robot embodiment, so it can be acted on.

  4. Verify

    Every result is checked against physical constraints before it is trusted.

  5. Return

    Verified, embodiment‑ready data goes back to a real robot — and the loop closes.

Nothing in the loop is simulated. What comes out is what a real robot can actually learn from.

Complete, deployable capability — not isolated components.

We work with partners to deliver working systems for defined vertical scenarios. Everything that leaves JOYX meets three tests.

Verified
Checked against the physics of the real scene, not assumed to be right.
Metric‑accurate
True to scale, so a robot can act on it without guesswork.
Embodiment‑ready
Expressed for the specific robot it is meant for.
“Raw video of the real world is abundant and nearly free; verified, metric‑accurate understanding built from it is scarce and valuable. That is what JOYX builds — and Hitch Interactive gives us the interfaces and the fleet to prove it where it matters: working systems in real vertical scenarios.”

One foundation engine. Two applications in the physical world.

The foundation engine that reads a live broadcast in real time is the same engine that reads the physical world for robots.

Physical AI division

Spatial intelligence for robots

Real‑to‑real robot training data and deployable capability for vertical scenarios, developed with robotics partners on real hardware.

JoyAd division

Spatial AI, live in broadcast sports

Broadcast‑grade virtual billboards inserted into live sports in real time, from the streaming signal alone. In production today.

JoyAd: monetize the blank.

Live sports command billions of engaged minutes, yet vast stretches of every frame go unmonetized. JoyAd finds the unused surfaces inside live video and fills them with broadcast‑grade billboards — generated, rendered and placed from a pure streaming signal. No on‑site hardware. No disruption to the broadcast.

One spatial engine, tuned across disciplines: golf, basketball, soccer, rowing, water polo, hockey, tennis and more.

“The future of sports media lives in the cloud — where technology drives both fan engagement and business growth. By transforming underutilized spaces into dynamic brand touchpoints directly within the live action, our zero‑CAPEX platform creates a true win‑win ecosystem.”
Tong Zhang, Founder and CEO, JOYX Inc. — on the multi‑year strategic partnership with Overnght, covering USRowing, USA Water Polo and Olympic, collegiate and emerging sports

Built on AWS. Ready for the world.

JOYX runs entirely on Amazon Web Services and has completed the AWS Foundational Technical Review — meeting AWS’s standards for technical, operational and security excellence.

AWS Foundational Technical Review — completed

Partners.

Newsroom.

About JOYX

JOYX Inc. is a U.S. Physical AI company that builds spatial intelligence — metric, three‑dimensional understanding of the physical world — from ordinary video. Its Physical AI division turns real‑world video into verified, metric‑accurate, embodiment‑ready robot training data, and its JoyAd division delivers AI‑powered virtual advertising insertion into live sports broadcasts.

Let’s talk.

Bringing a robot to a real job, or a brand to a live broadcast — start with an email.

info@myjoyx.com