A child interacting with the OriginX wheeled dual-arm robot at home
OriginX AI

For the happiness of every family

A physical-AI company built for the home. Our in-house world model keeps the robot evolving inside real families.

SCROLL
Wheeled dual-arm robot P1
Home · Full model

Wheeled dual-arm robot P1

A home robot that moves and manipulates with two arms. First-generation skills: conversation, tidying the desk, playing Gomoku.

Target under ¥20,000Child-safe by designMonthly OTA
FormOmni-directional base + dual arms + head RGB-D camera
DoF6 × 2 arms + 2 grippers; 3 on the base
Safety≤ 500 g per arm · ≤ 0.5 m/s at the tool · collision stop
Gen-1 skillsConversation · desk tidying · Gomoku
Desktop · Experiment model

MicroDuck

A desktop bipedal duck. It walks, sees and makes faces; children run balance and gait experiments on it and teach it motions by hand.

Target ¥3,000 classRL gaitMotion imitation · teleop
FormBipedal duck with head camera and an opening beak
DoF14 (5 × 2 legs + 4 neck/head)
LearningRL gait · motion imitation · phone teleoperation
SharedSame world model, curriculum and OTA pipeline as P1
Technology

Three parts, one self-evolving robot

The interactive operation platform carries human-robot interaction and skill delivery; SuperBrain evaluates and corrects the world model; the body brings that intelligence into the home. Task generation, training, evaluation and real-robot regression run automatically — humans only approve.

01

Interactive operation platform

Voice interaction, teleoperation, motion replay and synchronized capture. Skills land here — and so does the data.

02

SuperBrain

Evaluation and correction for the world model: imagine before acting, score and rank candidate actions, block unsafe ones, and fold real-robot outcomes back in.

03

The body

Brings intelligence into the home. Affordable, child-safe, updated over the air every month.

Home scene used every day Operation platform interaction · skills teleop · capture Real-robot data success / failure / fix opt-in only SuperBrain world-model eval & correction imagine, then act The body monthly OTA better every month The flywheel: model · real-robot data · body · scene
SuperBrain automatic-experiment console
SuperBrain · automatic experiment loop (prototype live)
Interactive operation platform console
Interactive operation platform · teleoperation and data capture (live)
About us

Bringing truly intelligent robots into every home

OriginX AI is a physical-AI company focused on the home. With our in-house world model at the center, we build a complete, self-evolving ecosystem around four elements — model, real-robot data, body and scene. The interactive operation platform carries human-robot interaction and skill delivery, SuperBrain handles world-model evaluation and correction, and the body brings that intelligence into the home. Together they fuse model, data and hardware into a business flywheel that gets stronger with use.

The home is the most complex and most demanding setting for embodied robots. We want a disruptive experience that brings genuinely intelligent robots into ordinary households and adds to the happiness of every family. Our core team comes from Rice, Cornell, UC Irvine and Fudan, with experience at NASA and Lightwheel, and brings leading technical depth in physical AI and world models together with hands-on industrial delivery.

Model

Our own world model: imagine, then act.

Real-robot data

Successes, failures and corrections from real homes.

Body

A home robot families can afford and children can use.

Scene

Starting with growing up alongside a child.

Team

Four people, four lanes

Technology / commercialization / marketing and fundraising / safety — each with an owner.

Qinzhen Ma
Qinzhen Ma
Founder / CEO · Technology

Head of Shanghai hardware & simulation and World Model Research Leader at Lightwheel; Rice Kavraki Lab, NASA internship, former senior algorithm engineer at Baidu.

Mingwei Sun
Mingwei Sun
Co-founder · Commercialization

12 years in cloud and AI industrialization; supply chain, channels and delivery at scale.

Gerald Xu
Gerald Xu
Partner · Marketing & fundraising

Cornell ECE MEng; employee #2 at Jingya Capital, backed eSign and Zilliz; co-founder of Vland, funded by Sequoia and Shunwei.

Shichen Tang
Shichen Tang
Co-founder · Safety gate

PhD in Mathematics, UC Irvine · MCS, Rice; optimization, planning, control and real-robot reliability testing.

Contact

Put robots in every home, with us

Partnerships, supply chain, investment — or if you want to join us — reach out directly.

WeChat
CareersEmail your CV and the role you want