As artificial intelligence shifts from virtual computing to physical interaction, embodied intelligence is emerging as a core force reshaping global productivity patterns. This technological revolution, bridging the digital and physical worlds, presents a multi-billion dollar market opportunity while facing significant bottlenecks such as costly data acquisition, the difficult Sim2Real gap, and a lack of industry standards. ZTE Corporation, a veteran communications equipment provider, is leveraging decades of expertise in the ICT sector to address these challenges. Centered on a "connectivity + computing power" strategy, the company has built a comprehensive technology stack covering "chip, soul, body, brain, cloud, and network." This approach not only tackles core obstacles to industrial-scale deployment but also aims to define a new development path for embodied intelligence, moving from technical exploration to widespread, accessible application.
The strategic value of embodied intelligence now surpasses that of smartphones and new energy vehicles, positioning it as the next epoch-defining industrial focus. Authoritative forecasts predict labor shortages reaching millions to tens of millions across major economies like China, the EU, and the US by 2030. On the policy front, China has included embodied intelligence in its list of key future industries for cultivation. The "Guiding Opinions on the Innovative Development of Humanoid Robots" targets the establishment of an internationally competitive industrial ecosystem by 2027. Cities including Beijing and Shanghai are planning trillion-yuan industrial clusters. Explosive market growth corroborates this potential. IDC data indicates a projected 508% year-on-year surge in global robot shipments by 2025. China's embodied intelligent robot market is expected to reach $77 billion. By 2026, China's humanoid robot market size is forecast to double, with service and consumer robot manufacturers accounting for over 85% of global shipments. User spending is projected to exceed $11 billion, sustaining a high growth rate of nearly 120%. Concurrently, "simulation-first" has become an industry consensus. The global robot simulation software market surpassed $2.5 billion in 2025, with a compound annual growth rate exceeding 30%.
However, bottlenecks on the eve of industry explosion are equally prominent. The unit cost of a humanoid robot prototype exceeds 200,000 yuan. Data collection is time-consuming, expensive, and carries safety risks. Visual and physical discrepancies between traditional simulation and real-world scenarios lead to difficult model transfer, extending R&D cycles to several months. There is a lack of standardized evaluation systems for features like dual-arm coordination and dexterous hand manipulation. System barriers exist between different robot brands, making low-latency, high-reliability communication a sticking point for scalable application. These pain points collectively point to a core solution: building an end-to-end simulation foundation that supports zero-shot Sim2Real transfer. This is precisely the direction of ZTE's breakthrough.
At a critical stage where embodied intelligence evolves from "point technologies" to "system capabilities," ZTE is utilizing its fully self-developed capabilities in communications, computing, and AI to create a complete technical system encompassing "simulation foundation - cloud-edge-end collaboration - communication base." This innovative approach addresses industry pain points. ZTE has introduced the RealMirror end-to-end embodied VLA simulation platform, achieving a full-chain闭环 covering "scene modeling - data acquisition - model training - simulation inference - automated evaluation - zero-shot Sim2Real transfer." Related research papers have been accepted by the 2026 International Conference on Robotics and Automation (ICRA 2026), earning international academic recognition.
To tackle high data acquisition costs, RealMirror features an industry-leading, low-cost dual-channel teleoperation framework using gestures and controllers. This lightweight teleoperation system, built on consumer-grade PICO VR headsets, replaces professional motion capture equipment costing hundreds of thousands of yuan or Apple's Vision Pro priced at tens of thousands, reducing total chain costs by over 90%. Official data shows that through a four-level cascade filtering algorithm and WebXR communication framework optimization, end-to-end latency is 114ms lower than mainstream frameworks, achieving 90Hz high-frequency data transmission. Single-arm pick-and-place task trajectory collection averages just 7.83 seconds, enabling the collection of 1,200 high-quality trajectories within 4 hours. This allows even small and medium-sized research teams to access industrial-grade simulation data.
Regarding evaluation benchmarks, RealMirror has built the world's first open-source VLA evaluation benchmark focusing on the dual-arm coordination characteristics of humanoid robots. It covers five typical scenarios: kitchen cleaning, air fryer operation, assembly line sorting, pouring between cups, and can stacking, providing over 1,200 high-quality simulation trajectories. The assessed capabilities span core skill dimensions like dual-arm coordination, precision operation, and dynamic grasping. The platform supports automated evaluation of mainstream algorithms like ACT, Diffusion Policy, and SmolVLA. It innovatively proposes a dual-dimensional evaluation system based on "task success rate + skill capability distribution," promoting the evolution of algorithm research from "demo-driven" to "data-driven."
The most critical breakthrough lies in zero-shot Sim2Real transfer capability. ZTE has pioneered the systematic application of 3D Gaussian Splatting technology for robotic simulation environment reconstruction, proposing a dynamic-static hybrid 3D Gaussian digital twin solution. This enables differentiated high-fidelity reconstruction of static environments, the robot body, and interactive objects, increasing visual consistency between simulation and real scenes to over 95%. Experimental validation on the Zhiyuan A2 real robot showed that models trained in simulation achieved a 93% success rate in basic grasping tasks and a 71% success rate in complex dual-arm sphere transfer tasks without any fine-tuning on real data, breaking the traditional simulation-reality technical gap.
The MirrorLimb low-cost dual-channel access scheme, derived from RealMirror, further lowers the industry entry barrier. This solution supports deep coupling between consumer-grade PICO Neo3 Pro VR devices and the industrial-grade Isaac Sim simulation ecosystem, at a cost仅为 1/15th of an Apple Vision Pro-based solution. It establishes a普惠化 data collection paradigm of "thousand-yuan level equipment + industrial-grade quality."
ZTE has constructed a full-stack collaborative architecture comprising "cloud-based embodied intelligence hub - wireless edge embodied intelligence platform - device-side robot terminals." This breaks the limitations of single-unit intelligence and achieves a breakthrough in swarm intelligence for embodied intelligence, described by the industry as "redefining the systemic capability boundaries of embodied intelligence." The cloud intelligence hub integrates a vision-language model with hundreds of billions of parameters, enabling fully autonomous task planning. Combined with RAG technology to interface with enterprise knowledge bases, question-answering accuracy is improved by over 40%. Swarm intelligence coordination efficiency and task planning response speed are increased by 40% and 50% respectively. End-to-end natural interaction latency is under 1 second, achieving a leap from "passive execution" to "active decision-making."
The wireless edge embodied intelligence platform enhances the operational speed of robots like those from Zhiyuan and Zhuoyide by four times through a comprehensive motion acceleration scheme. For instance, Zhuoyide robot grasping speed improved from 23.2 seconds to 5-6 seconds, and Zhiyuan robots from 16.4 seconds to 3-4 seconds, approaching human operational levels. The platform incorporates speech emotion recognition technology for anthropomorphic action generation, increasing customer satisfaction by 35%. Users have评价 it as "elevating the service experience of embodied intelligent robots to a new level." On the device side, the architecture achieves deep integration of lightweight large model inference, multimodal perception, and whole-body motion control, realizing a full-chain闭环 of "thinking + skills + perception + motion control." This significantly enhances the scenario adaptability and operational precision of embodied intelligence, making it one of the most complete collaborative architectures available.
As a global leader in communications, ZTE is deeply integrating 5G-Advanced technology with embodied intelligence. The embodied intelligent wireless communication network it has built solves connectivity challenges for multi-robot coordination, serving as the "core communication guarantee for the scalable application of embodied intelligence." Based on ZTE's self-developed 5G-A technology, this network supports uRLLC, Massive MIMO, and intelligent beamforming. With embedded AI for intelligent wireless resource scheduling, it supports seamless robot mobility and can stably operate over 50 heterogeneous robots concurrently, achieving 99.99% network availability. ZTE's 5G-A EasyOn·Robot dedicated network solution productizes 5G-A network capabilities, offering robots "plug-and-play" wireless connectivity. It addresses pain points like Wi-Fi instability, inflexible wiring, and multi-robot interference, praised by partners as the "optimal solution for robot connectivity." At the 2025 World Robot Conference, ZTE partnered with Zhuoyide, Zhiyuan, and others to demonstrate the world's first 5G-A-enabled heterogeneous swarm intelligence collaborative guided tour system. It achieved unified scheduling of cross-vendor robots, improving task response speed by 50% with end-to-end interaction latency under 1 second, marking a step from "single-unit intelligence" to "swarm intelligence."
True industry leadership lies not only in technological innovation but also in promoting holistic industrial collaborative development. ZTE fosters a virtuous cycle of "technological innovation - ecological synergy - scenario deployment" through open-source ecosystem building, industry standard leadership, and breakthroughs in practical application. Beyond making the RealMirror platform fully open-source, ZTE has released an embodied VLA simulation R&D kit with out-of-the-box toolchains, supporting rapid deployment in five typical scenarios. It includes over 1,200 simulation trajectories and an automated evaluation system, significantly lowering the R&D barrier. University research teams have stated it "allows the entire industry to stand on a higher starting point for innovation."
In standard setting, ZTE is a core participant due to its technical积累, described as a "core force driving industry standardization." As a key member of bodies like MIIT-TC8 and the National Robotics Standardization Technical Committee, ZTE leads the drafting of standards and deeply participates in formulating multiple standards, promoting a unified, standardized system. Industry experts note that unified standards are a prerequisite for industrial scale-up, and ZTE plays a key leading role therein.
For practical deployment, ZTE focuses on "technological innovation + scenario adaptation," pushing technology from the lab to real-world use. Its practicality and replicability are highly recognized. In industrial manufacturing, ZTE has deployed embodied intelligent robots in its own 5G smart manufacturing factory, successfully replacing manual labor in precise insertion/extraction tasks for RJ45 connectors and optical modules. With a 1ms real-time control cycle, it has addressed the final breakpoint in fully automated production lines, hailed as a "breakthrough application for industrial flexible assembly." In exhibition guide scenarios, its "cloud-edge-end" collaborative solution has been successfully used at major events like MWC Shanghai, enabling proactive greeting and anthropomorphic explanation functions. Furthermore, the technological成果 can be widely applied in warehousing logistics, equipment inspection, elderly care, and other scenarios, demonstrating strong adaptability. ZTE also collaborates deeply with robot manufacturers like Zhuoyide and Zhiyuan, completing capability abstraction and protocol unification to enable plug-and-play multi-brand devices. This creates an ecological闭环 of "connectivity + computing + AI + terminals,"推动 the industry from "going it alone" to "ecological collaboration."
ZTE's leading position in the embodied intelligence industry stems from decades of technological accumulation in ICT, full-stack self-development capabilities, and a global industrial layout. As a leading global ICT infrastructure provider, ZTE ranks second globally in shipments of 5G base stations and core network equipment, and first in global market share for 5G FWA&MBB. In computing infrastructure, revenue from its computing power business grew 180% year-on-year in the first three quarters of 2025, accounting for 25% of total revenue. Its AI servers are used by leading domestic internet companies and widely applied in finance, power, and other sectors. Technologically, ZTE has built a moat: its subsidiary ZTE Microelectronics has independently developed and commercialized over 150 types of chips. Its self-developed DPU chip provides algorithm hardware acceleration and protocol offloading capabilities, solidifying the computing foundation for embodied intelligence. In AI, ZTE holds nearly 5,500 patent applications, with nearly half granted. It ranked second among China's Top 100 Innovative Enterprises in 2025 and was首次入选 the Clarivate Top 100 Global Innovators list in 2026, indicating its patents reach globally top levels in influence and technological uniqueness. The integrated innovation of "communications + computing power + AI + robotics" allows ZTE to leverage 5G-A/6G for hundreds of Mbps uplink bandwidth, millisecond-level deterministic low latency, and high-reliability connectivity, providing极致 network assurance for real-time robot command acquisition and cloud collaboration. This multi-faceted differentiated advantage is difficult for others to replicate. ZTE offers not just single products or technologies but, centered on "connectivity + computing power," achieves deep integration of communications, computing, AI, and robotics technologies. It has built a full-chain technology system from underlying chips to upper-layer applications, positioning itself as a builder of the industrial technology foundation and an enabler of the ecosystem.
Currently, the embodied intelligence industry is in a period of rapid development, driven by the dual engines of technological innovation and practical application, poised for a new wave of growth. ZTE has a clear development plan: short-term focus on breakthroughs for industrial applications, mid-term cost reduction for home market preparation, and a long-term vision of human-robot coexistence. From technological breakthroughs to ecosystem building, from standard leadership to scenario deployment, ZTE is leveraging its comprehensive advantages to lead the embodied intelligence industry from technical exploration towards large-scale, accessible application. As intelligent robots move from high-end labs to factory floors, urban spaces, and homes, ZTE's mission is becoming reality. In this technology revolution reshaping productivity patterns, ZTE is not just a participant and promoter but an innovative force defining a new paradigm for industrial development, ultimately contributing to a more efficient, safer, and more convenient intelligent life for humanity.
Comments