Deployment Era Starts: How One Firm Delivered Seven Turnkey Embodied‑AI Solutions Without Selling Robots

Huang Renxun once said that in the AI era, tokens are the currency, and embodied agents will become the biggest token consumers; ZhiYuan’s co‑founder Peng Zhihui identifies embodied intelligence as the answer.

Embodied intelligence’s task space spans the digital and physical worlds, where a robot continuously perceives, reasons, decides and controls, creating an endless flow of tokens that fuels a new production‑infrastructure.

At the Shanghai conference on 17 April, ZhiYuan unveiled four new robot bodies, six AI models and seven standardized productivity solutions, together with a full‑stack ecosystem called AIMA, declaring the start of the embodied‑AI deployment era.

"One Body Three Intelligence" Architecture

The "One Body" refers to a stable, high‑capacity, low‑cost robot chassis, while the "Three Intelligences" cover motion intelligence, interaction intelligence and operation intelligence, jointly forming the physical and cognitive foundation for general‑purpose robots.

Four New Robot Bodies

Yuanzheng A3 : a full‑size humanoid with the industry’s highest thrust‑to‑weight ratio, modular structure, hot‑swappable dual batteries, 8‑10 h continuous operation, and the first humanoid to natively support deployment‑state interaction intelligence.

Lingxi X3 : a new‑modal humanoid service terminal that combines intelligence, anthropomorphism and safety, aimed at service‑industry tasks, slated for release by year‑end.

G2 Series : two wheeled robots – G2 Air , a lightweight platform supporting human‑machine unified data acquisition, and G2 Max , an industrial‑grade carrier with higher payload.

Kuotu D2 Series : a quadruped robot extending embodied intelligence to all‑terrain scenarios, featuring adaptive dynamic balance and L3‑level autonomous driving for security patrol, fire‑rescue and other special operations.

Six AI Models

Behavior Foundation Model (BFM) : trained on over 100 million frames and 700 hours of motion‑capture data, enabling high‑precision human‑like motion imitation and zero‑shot task adaptation; demonstrated with G2 Air in a "shadow mode" where a human guides the robot and data is generated automatically.

Generative Control Model (GCFM) : a first‑in‑industry generative control model that creates arbitrary motions in real time from natural‑language or audio commands, eliminating the need for pre‑programmed sequences.

WITA Omni 1.0 : an end‑to‑end embodied multimodal interaction model built on the WITA large model, preserving emotion, context and tone, supporting mid‑utterance interruption and correction, with interaction latency under 500 ms.

GO‑2 and upcoming GO‑3 : models that fuse a "large brain" and a "small brain" to give robots reasoning, planning and long‑term task execution capabilities; GO‑3 will scale data volume to tens‑hundreds of times that of GO‑2.

Action World Model (GE‑2) and the open dataset AGIBOT WORLD 2026 , together with the simulation platform Genie Sim 3.0 and development studio Genie Studio 2.0 .

AIMA Full‑Stack Ecosystem

The AI Machine Architecture (AIMA) ecosystem adopts a "1+3+X" structure: the open‑source operating system Link‑U OS , three core development platforms (the motion‑learning platform Ling‑Chuang, the agent‑customization platform Ling‑Xin, and the low‑threshold one‑stop platform Genie Studio), and a set of embodied‑agent frameworks.

To address the data bottleneck of embodied model training, ZhiYuan launched the world’s first physical‑AI data network "Hive Data Co‑creation Action", targeting millions of hours of data production within a year and a hundred‑billion‑hour scale by 2030.

Industrial Deployment Cases

In a 3C precision‑manufacturing line at Longqi’s tablet factory, two G2 robots acted as regular staff, handling pick‑and‑place of tablets, achieving an 18‑20 s action cycle, 100 % success rate and throughput of over 300 items per hour; ZhiYuan plans to scale to hundreds of units by Q3 2025 and has already deployed robots at Intel, Juneng, and other factories.

Other scenarios include industrial搬运 (G2 Max), logistics sorting (flexible item and soft‑package handling), retail guide and service stations (partnering with Anta, Haidilao, etc.), security patrol (quadruped robots, VLN model) and commercial cleaning (targeting 6 000 units this year with a cumulative 10 000 units shipped).

All seven scenarios are supported by standardized solutions co‑created with leading industry customers, and a global robot‑rental platform "Qingtian Rental" enables a RaaS model with on‑demand usage and value‑based pricing.

Market Impact and Growth

In March 2026 ZhiYuan announced the mass‑production of the first 10 000 general‑purpose embodied robots. In 2025 the company shipped over 5 100 units, capturing 39 % of the global market and achieving the "double first" in volume and share.

Revenue grew from ¥300 k in 2023 to ¥60 M in 2024 and surpassed ¥1 B in 2025, making it the fastest Chinese robot firm to cross the ¥1 B threshold.

Future Development Curves

ZhiYuan outlines three curves: the past X‑curve (development‑trial phase) that moved robots from prototype to mass‑production; the upcoming Y‑curve (2026‑2030 deployment growth) where robots autonomously execute tasks, maintain themselves and generate value, driving a data‑flywheel; and the post‑2030 Z‑curve (deployment‑ubiquity) likened to a "ChatGPT 3.5 moment" where robots pass a physical Turing test and achieve superhuman productivity across manufacturing, logistics and services.

Conclusion

By covering most demand scenarios with a full product matrix, backing hardware with powerful AI models, offering standardized deployment solutions that cut costs, and building a global ecosystem, ZhiYuan has closed the commercial loop for embodied intelligence. The scaling of mass production will feed data accumulation, accelerating the emergence of truly autonomous, productive robots.