CtrlK
| Xiaomi Robotics Division | |
|---|---|
| General information | |
| Parent company | Xiaomi Corporation |
| Headquarters | Beijing, China |
| Founded | 2021 (Robotics Lab) |
| Key person | Lei Jun (Founder & CEO) |
| Industry | Consumer electronics, robotics, electric vehicles |
| Notable products | CyberDog, CyberOne, Xiaomi-Robotics-0 |
| Website | mi.com |
Xiaomi Corporation is a Chinese multinational technology company that has expanded aggressively into robotics and embodied intelligence since establishing its dedicated Robotics Lab in August 2021. Best known as one of the world's largest smartphone manufacturers and consumer electronics companies, Xiaomi has applied its expertise in hardware engineering, artificial intelligence, mass manufacturing, and smart ecosystem integration to develop a series of increasingly sophisticated robots. These range from the open-source CyberDog quadruped (2021) to the full-size CyberOne bipedal humanoid robot (2022), and most recently to factory-deployed humanoid systems achieving a 90.2% task success rate in Xiaomi's electric vehicle plant (2026).
Xiaomi's robotics program is distinguished by its integration with the company's broader "Human x Car x Home" ecosystem strategy, its commitment to open-source AI models, and CEO Lei Jun's stated vision that humanoid robots will become central to both industrial manufacturing and household life within the next decade.
Xiaomi was founded in April 2010 by Lei Jun in Beijing. The company initially focused on smartphones, rapidly growing into one of the world's top three smartphone manufacturers by shipping volume. Over time, Xiaomi diversified into a vast ecosystem of connected products, including smart home devices, wearables, laptops, televisions, and Internet of Things (IoT) appliances. By 2024, Xiaomi's AIoT (AI + IoT) platform connected over 500 million smart devices worldwide through its XiaoAI voice assistant and Mi Home platform.
In 2024, Xiaomi entered the electric vehicle market with the launch of the Xiaomi SU7 sedan, produced at a state-of-the-art factory in the Beijing Economic-Technological Development Area spanning 720,000 square meters. The factory achieves a production cycle of 76 seconds per vehicle, with over 700 industrial robots operating around the clock and a 91% overall automation rate in key workshops. By mid-2025, Xiaomi had delivered over 250,000 vehicles. In its fiscal year 2025, Xiaomi's total revenue exceeded 457 billion yuan (approximately $63 billion), a 25% year-on-year increase, with adjusted net profit of 39.2 billion yuan.
This combination of massive-scale hardware manufacturing, vertically integrated supply chains, advanced AI capabilities, and a connected device ecosystem provides the foundation for Xiaomi's robotics ambitions.
Xiaomi formally established its Robotics Lab in August 2021, signaling a strategic commitment to robotics research and development. The lab's research spans several focus areas:
| Research area | Description |
|---|---|
| Bionic perception and cognition | Emotion recognition, environmental sound classification, spatial understanding |
| Biomechatronics | Custom actuators, servo motor design, tendon-driven mechanisms |
| Artificial intelligence | Vision-language-action models, reinforcement learning, tactile perception |
| Computer vision | Depth sensing, 3D spatial recognition, object detection |
| Locomotion control | Bipedal walking, quadruped gaits, whole-body coordination |
As of late 2025, Xiaomi's robotics division had over 250 open job positions across engineering roles, with a particular focus on dexterous manipulation. The company has been actively recruiting top-tier talent, including hiring Lu Zeyu, a former member of Tesla's Optimus dexterous hand team, who joined Xiaomi in late 2025 to lead dexterous hand R&D. During his tenure at Tesla, Lu Zeyu had been deeply involved in developing the Optimus robot's tactile sensors, dexterous grasping, and hand structure design.
The Robotics Lab's first product was CyberDog, an open-source bionic quadruped robot launched on August 10, 2021. Inspired by platforms like Boston Dynamics' Spot, CyberDog was designed as an accessible, developer-friendly robot at a fraction of the cost of comparable platforms.
| Specification | Value |
|---|---|
| Type | Quadruped robot |
| Weight | ~3 kg |
| Processor | NVIDIA Jetson Xavier NX (384 CUDA cores, 48 Tensor cores) |
| Degrees of freedom | 12 (via custom high-torque servo motors) |
| Maximum speed | 3.2 m/s (~7 mph) |
| Sensors | 11 total: Intel RealSense D455 depth camera, 6 microphones, touch sensors |
| Connectivity | USB-C (x3), HDMI, WiFi, Bluetooth |
| Software | Open-source (system code and mechanical blueprints) |
| Price | 9,999 yuan (~$1,540) |
| Units produced | 1,000 (initial release to engineers, researchers, fans) |
CyberDog's capabilities included autonomous navigation, object tracking, SLAM (simultaneous localization and mapping), obstacle avoidance, face recognition, voice recognition, and dynamic movements such as backflips. Xiaomi released the robot's system code and mechanical blueprints through the Xiaomi Open Source Community, fostering a collaborative developer ecosystem.
The project served a dual purpose: it provided Xiaomi with foundational experience in servo motor design, AI-driven locomotion control, and sensor integration, while also demonstrating the company's ability to deliver robotics hardware at consumer-friendly price points.
Xiaomi released CyberDog 2 in August 2023, a significant upgrade designed to resemble a Doberman. At 8.9 kg with a height of 36.7 cm, the second-generation robot featured proprietary CyberGear micro-actuators, 19 high-precision sensors (including LiDAR and Intel RealSense D430), a maximum speed of 1.6 m/s, and integration with Xiaomi's XiaoAI smart home platform. Priced at 12,999 yuan (~$1,790), CyberDog 2 ran Ubuntu 18.04 with ROS 2 for developer accessibility.
CyberOne is a full-size bipedal humanoid robot unveiled on August 11, 2022, at a Xiaomi product launch event in Beijing. The robot was introduced by Lei Jun himself, who received a flower from CyberOne on stage in a demonstration that was widely covered in technology media. Standing 177 cm tall and weighing 52 kg, CyberOne represented Xiaomi's leap from quadruped to bipedal humanoid robotics.
| Specification | Value |
|---|---|
| Height | 177 cm (5 ft 10 in) |
| Weight | 52 kg (115 lb) |
| Arm span | 168 cm |
| Degrees of freedom | 21 |
| Maximum walking speed | 3.6 km/h (2.2 mph) |
| Single-hand payload | 1.5 kg |
| Joint response time | 0.5 milliseconds |
| Upper limb motor weight | 500 g (30 Nm rated torque, 96 Nm/kg power density) |
| Hip joint peak torque | 300 Nm |
| Vision system | Mi-Sense depth module + Intel RealSense D455 |
| Emotion recognition | 45 human emotional states (MiAI Vocal Emotion Engine) |
| Sound classification | 85 environmental sound types (MiAI Environment Semantics Engine) |
| Face display | Curved OLED module |
| Production cost | ~600,000 to 700,000 yuan ($89,000 to $104,000) per unit |
CyberOne's design philosophy prioritized human-robot interaction and emotional intelligence rather than athletic performance. The robot's proprietary MiAI Vocal Emotion Identification Engine classifies a speaker's emotional state into one of 45 categories by analyzing tone, pitch, and cadence. The MiAI Environment Semantics Recognition Engine identifies 85 distinct types of ambient sound for situational awareness. Both engines were developed entirely in-house at the Xiaomi Robotics Lab.
The unveiling came just weeks before Tesla's AI Day on September 30, 2022, where Elon Musk presented the first prototype of Tesla Optimus. At the time of their respective debuts, CyberOne demonstrated more polished walking and object manipulation capabilities than Tesla's initial prototype, although Tesla's program has since advanced rapidly through multiple generations.
CyberOne was subsequently showcased at Mobile World Congress (MWC) 2023 in Barcelona, bringing the robot to an international audience.
Xiaomi has developed multiple AI models and software frameworks to power its robotics program, with a strong emphasis on open-source release.
In November 2025, Xiaomi released and open-sourced MiMo-Embodied, described as the industry's first cross-domain foundation model integrating autonomous driving and embodied AI. The 7-billion-parameter vision-language model achieved state-of-the-art performance across 17 embodied AI benchmarks (covering task planning, affordance prediction, and spatial understanding) and 12 autonomous driving benchmarks (covering environmental perception, status prediction, and driving planning). Xiaomi planned to deploy MiMo-Embodied across its robot vacuum cleaners, factory AGVs, and SU7 intelligent driving systems starting in Q1 2026.
On February 12, 2026, Xiaomi announced Xiaomi-Robotics-0, a 4.7-billion-parameter open-source Vision-Language-Action (VLA) model designed for real-time robotic execution. The model represents Xiaomi's first-generation robot large-scale model, combining visual perception, natural language understanding, and direct robotic control in a single architecture.
| Aspect | Detail |
|---|---|
| Parameters | 4.7 billion |
| Architecture | Mixture-of-Transformers (MoT): decoupled VLM + multi-layer Diffusion Transformer (DiT) |
| VLM role | Handles ambiguous instructions and spatial reasoning |
| DiT role | Uses flow matching to generate high-frequency continuous action chunks |
| Training data | Large-scale cross-embodiment robot trajectories + vision-language data |
| Key innovation | Timestep alignment of consecutive action chunks for seamless real-time rollouts |
| Deployment | Consumer-grade GPU |
| LIBERO benchmark | 98.7% average success rate |
| SimplerEnv Visual Matching | 85.5% |
| CALVIN benchmark | Strong results (state-of-the-art across all simulation benchmarks) |
| Real-robot validation | Dual-arm bimanual manipulation tasks (towel folding, block disassembly) |
| Availability | Open-source (code, model weights, documentation on GitHub and Hugging Face) |
The architecture decouples the Vision-Language Model from the action generation component. The VLM processes visual inputs and language instructions, handling ambiguous commands and spatial reasoning. The Diffusion Transformer then uses flow matching to generate high-frequency, continuous action sequences. A post-training technique aligns consecutive predicted action chunks to enable smooth, real-time execution on physical robots.
TacRefineNet is Xiaomi's proprietary tactile perception and fine-tuning model, designed to enable robotic manipulation guided primarily by touch rather than vision alone. The system processes haptic feedback from tactile sensors across a robot's hands, allowing it to adapt to physical variables in real time. In factory environments where visual perception can be unreliable due to lighting changes or occlusion, TacRefineNet provides a complementary sensory channel.
Xiaomi open-sourced the TacRefineNet framework along with 61 hours of raw tactile data collected through sensorized tactile gloves worn by human operators, supporting broader research in robotic manipulation.
In early March 2026, Xiaomi announced that humanoid robots derived from the CyberOne platform had begun performing real production tasks at its electric vehicle manufacturing facility in Beijing. Xiaomi President Lu Weibing described the robots as "interns" on the production line, marking one of the first instances of a consumer technology company deploying its own humanoid robots in its own factory.
The robots were stationed at a self-tapping nut assembly workstation in the die-casting workshop of Xiaomi's EV factory. The task required the robot to:
| Metric | Result |
|---|---|
| Task | Self-tapping nut installation (dual-side simultaneous) |
| Success rate | 90.2% |
| Cycle time | 76 seconds (met production line requirement) |
| Continuous autonomous operation | 3 hours |
| Control approach | End-to-end data-driven with reinforcement learning |
| Sensory integration | Vision + tactile feedback + joint proprioception |
| Motion control | Hybrid architecture (optimization-based control + reinforcement learning) |
| Optimization controller speed | Sub-1-millisecond iterations |
The primary technical challenge involved precise alignment and reliable engagement of self-tapping nuts, complicated by spline structures, non-fixed gripping positions, and magnetic interference. The hybrid motion control architecture combined optimization-based controllers running at sub-millisecond iteration speeds with reinforcement learning policies trained on multimodal sensory data.
The robots were also validated across additional workstations, including bin-picking tasks and front badge installation on vehicles.
The 90.2% success rate, while not yet matching the reliability required for fully autonomous factory operation, demonstrated that Xiaomi's humanoid platform could perform meaningful industrial work. The 3-hour continuous operation window and the ability to meet the production line's 76-second cycle time requirement showed practical viability for specific workstations. Lei Jun stated that Xiaomi plans to deploy "a large number" of humanoid robots across its factories within five years.
In April 2026, Xiaomi unveiled a significantly upgraded bionic hand for the CyberOne platform, introducing several innovations designed to address longstanding challenges in robotic manipulation and thermal management.
The new hand achieved a 60% volume reduction compared to the previous version, shrinking from 228 x 105 x 64 mm to 187 x 88 x 36 mm. This brings the hand to a 1:1 proportional match with an average adult human hand (based on a 1.73 m body model). The redesign features a 50% increase in total degrees of freedom and an 83% jump in active degrees of freedom, bringing the hand into the 22 to 27 DOF range necessary to replicate authentic human hand kinematics.
Xiaomi expanded tactile sensor coverage to 8,200 square millimeters across the entire palm surface, encompassing fingertips, finger pads, and the palm. This full-palm sensing allows the hand to perform manipulation tasks effectively even when vision systems are occluded or unreliable. Training data for the tactile system was collected using sensorized tactile gloves worn by human operators, feeding into the dual-model software stack of Xiaomi-Robotics-0 (for visual-language-action processing) and TacRefineNet (for tactile refinement).
The most novel feature is an evaporative cooling system that mimics human perspiration. The hand integrates 3D-printed metal liquid cooling channels throughout its structure, connected to a micro-pump. The system works by circulating liquid coolant through these channels, where it evaporates at the surface at a rate of 0.5 milliliters of water per minute.
| Thermal management specification | Value |
|---|---|
| Cooling mechanism | Evaporative liquid cooling (biomimetic "sweating") |
| Channel fabrication | 3D-printed metal |
| Evaporation rate | 0.5 mL/min |
| Heat dissipation | ~10 W removed from 100 W motor load |
| Purpose | Prevent thermal throttling during sustained high-load operation |
A single robotic hand performing heavy-duty tasks can draw over 100 watts of power. At approximately 70% efficiency, more than 30 watts are converted directly into heat, and under stall conditions (when the motor is under load but not moving), this thermal load limits continuous operation time. Traditional tendon-driven robotic hands typically fail after roughly 10,000 grasping cycles due in part to thermal degradation. Xiaomi's sweating mechanism enabled the hand to survive over 150,000 continuous grasping cycles without failure, a 15-fold improvement that fundamentally changes the economics of factory deployment.
Lei Jun has been the primary public champion of Xiaomi's robotics program, articulating a long-term vision that positions humanoid robots alongside smartphones and electric vehicles as transformative product categories.
At the 2022 CyberOne unveiling, Lei Jun stated that "CyberOne's AI and mechanical capabilities are all self-developed by Xiaomi Robotics Lab," and described intelligent robots as future companions and assistants in everyday life.
Following the 2026 factory trials, Lei Jun emphasized that humanoid robots would be "central to how factories operate" within five years, and that household applications could "open a new trillion-yuan market." He framed the current state of humanoid robotics as analogous to the early days of the smartphone industry, where rapid cost reduction and capability improvement would eventually bring robots into mainstream adoption.
At the 2026 Two Sessions (China's annual legislative meetings), Lei Jun submitted five formal proposals as a National People's Congress representative. Two of these focused specifically on robotics:
Standardization and reliability: Lei Jun proposed that by 2027, the Mean Time Between Failures (MTBF) of humanoid robots in industrial scenarios should exceed 10,000 hours. Operating 20 hours per day, this would mean roughly 500 days of continuous operation without failure, which he characterized as the entry threshold for a robot to qualify as a "formal employee" rather than an "apprentice."
International standards alignment: He advocated for aligning domestic Chinese robotics standards with international standards to help Chinese humanoid robots expand into global markets.
Lei Jun acknowledged that humanoid robots remain at a critical transition point from laboratory to industrialization, with bottlenecks including poor process stability, high hardware costs, and limited application scenarios. He proposed expanding smart manufacturing application scenarios, introducing industrial subsidies and policy incentives, and strengthening safety standards.
Xiaomi operates within an intensely competitive Chinese humanoid robotics industry that has seen rapid growth and increasing consolidation since 2024.
According to the Beijing Academy of Artificial Intelligence, China's humanoid robotics companies are organized into three tiers:
| Tier | Companies |
|---|---|
| First tier | Unitree Robotics, Zhiyuan Robotics, UBTECH Robotics, Galbot |
| Second tier | Galaxea Dynamics, LimX Dynamics, Xiaomi, Meituan |
| Third tier | Casbot, SenseTime, XPeng |
Xiaomi's placement in the second tier reflects the fact that, unlike pure-play robotics companies, Xiaomi treats robotics as one division within a much larger technology conglomerate. However, analysts have noted that if Xiaomi decides to fully commit its manufacturing scale and resources, it could reshape the competitive landscape significantly.
| Company | Key robot | Approach | 2025-2026 shipments |
|---|---|---|---|
| Unitree Robotics | H1, G1 | Pure-play robotics; consumer and industrial | 10,000-20,000 units forecast (2026) |
| UBTECH Robotics | Walker S series | Industrial logistics and manufacturing | ~1,000 units (2025) |
| Tesla | Optimus | Integrated with EV manufacturing and AI infrastructure | Pre-production testing |
| Xiaomi | CyberOne | Integrated with EV factory, AIoT ecosystem | Factory "intern" trials |
| Agibot | A2 | Industrial manufacturing focus | Among top Chinese shippers |
Xiaomi's primary competitive advantages include its massive manufacturing infrastructure, vertical integration across hardware and software, a connected ecosystem of over 500 million devices, and synergies with its electric vehicle division, which provides both a testbed for deployment and shared technology in autonomous navigation, sensor fusion, and AI computing. The company's willingness to open-source key AI models (Xiaomi-Robotics-0, TacRefineNet, MiMo-Embodied) also differentiates its approach by fostering community-driven development.
TrendForce forecasts 2026 as a pivotal year for humanoid robot commercialization, with global shipments expected to exceed 50,000 units (over 700% year-on-year growth). Morgan Stanley projects the humanoid robot market could reach $5 trillion by 2050. China's 15th Five-Year Plan, announced at the 2026 Two Sessions, prioritizes AI and robotics as strategic industries. On March 3, 2026, Chinese authorities released the country's first national standard system for humanoid robots and embodied intelligence, covering foundational standards, computing, limbs and components, full-system integration, applications, safety, and ethics.
Xiaomi's robotics program fits within a deliberate, step-by-step corporate strategy of building capabilities across connected product categories.
| Year | Milestone |
|---|---|
| 2010 | Xiaomi founded; smartphone focus |
| ~2016 | Entry into robotic vacuum cleaners and IoT devices |
| 2021 | Robotics Lab established; CyberDog quadruped released |
| 2022 | CyberOne humanoid unveiled |
| 2023 | CyberDog 2 released; CyberOne shown at MWC Barcelona |
| 2024 | Xiaomi SU7 electric vehicle launched; factory operational |
| 2025 | MiMo-Embodied cross-domain AI model released; talent recruitment (ex-Tesla engineers) |
| 2026 | Xiaomi-Robotics-0 VLA model; factory deployment (90.2% success); bionic hand with thermal sweating |
| ~2031 (target) | Large-scale humanoid robot deployment across Xiaomi factories |
The progression from robotic vacuums to CyberDog to CyberOne to factory-ready humanoid robots reflects a company systematically building expertise in actuator design, locomotion control, perception, and AI. Each product generation has served as a stepping stone, with technologies developed for one platform informing the next. The EV factory provides a controlled environment for validating humanoid robots before any potential expansion into the far more complex and unpredictable household domain.