At the 2026 World Artificial Intelligence Conference, Pudu Robotics unveiled its first human-like intelligent robot, the PUDU D7, marking its official transition from specialized robots to general-purpose embodied intelligence. Pudu has established four major product lines—service and delivery, commercial cleaning, industrial logistics, and embodied intelligence—covering 16 scenarios. It leads globally in commercial service robot revenue, shipment volume, and international expansion, with over 130,000 units delivered worldwide. At the core technology level, Pudu has developed a “one brain, multiple forms” architecture, enabling a unified intelligent brain to adapt to various configurations including specialized, human-like, and humanoid robots. Leveraging 50 million hours of real-world operational data generated annually by its global fleet of 130,000 robots, Pudu has created a true-world data flywheel. In April, the company completed nearly RMB 1 billion in funding, achieving a valuation exceeding RMB 10 billion, and is accelerating the evolution of robots from “execution” to “cognition.”Article author and source: 36Kr
Embodied intelligence has entered the second half, where winners will no longer be determined solely by robots. On July 17, the annual World Artificial Intelligence Conference opened in Shanghai, where the heat was not just in the weather but also in the tangible and intangible competition within the exhibition halls.
Among them, the embodied intelligence exhibition area was particularly intense, with reports indicating that 208 embodied intelligent terminals and over 300 physical units would be showcased. Amid a variety of differently shaped robots, a robot with a dark gray matte finish stood steadily at the forefront of the crowd. Composed of large square geometric blocks, it featured smooth, gentle curves at every joint and connection point. Its four-legged wheeled chassis enabled it to glide smoothly and stably through the complex exhibition aisles.
This is the first in-person unveiling of the PUDU D7, a humanoid intelligent robot by Pudu Robotics, designed for industrial, retail, and other scenarios.
Over the past two years, the embodied intelligence industry has experienced rapid growth. Market feedback has been straightforward: according to IT Juzi’s event database, domestic funding in the embodied intelligence sector reached RMB 93.5 billion in the first half of 2026, a fivefold increase compared to the first half of 2025.
Large amounts of capital, talent, and attention have poured in, while debates over a bubble continue. Beneath the noise, some in the industry view 2026 as the pivotal year for mass production of humanoid robots, marking a critical milestone for commercialization and real-world validation. After two years of demonstrations, what the market truly needs—and is waiting for—is embodied AI technology that can be deployed in real-world scenarios to solve actual problems.
The turning point has arrived, regardless of whether companies are ready. Those who can most quickly deploy robots in factories, supermarkets, and hospitals will be the winners. Behind this lies not only R&D capability in the lab, but also challenges in strategy, technology, product development, commercialization, supply chain, and organization.
As a global leader in commercial robotics, Pudu Robotics began a decade ago with food delivery in the restaurant industry and now serves multiple sectors including retail, hospitality, industrial, warehousing and logistics, food service, property cleaning, healthcare, entertainment and sports, education, public transportation, and services. With cumulative shipments exceeding 130,000 units and a valuation in the billions, Pudu is directly competing in this pivotal battle for embodied AI deployment through its systematic approach of in-house hardware and software development, coupled with a highly efficient data and business feedback loop.
01 Expanding a diverse range of products across all categories to establish dominance in real-world scenarios. “Unlike the C-end product, where failing to win initially makes it hard to recover, B-end business is a multi-round博弈 that requires endurance rather than explosive growth, allowing companies to trade time for space through diversified product offerings.” In May, Zhang Tao, founder and CEO of Pudu Robotics, stated this during his speech at BEYOND Expo in Macau.
Pudu has executed this strategy with determination and success: expanding from service delivery robots to commercial cleaning robots, industrial delivery robots, and finally general-purpose embodied intelligent robots, with these four highly diverse product lines supporting one another.
In its earliest adopted service delivery product line, in addition to maintaining global market leadership, cumulative revenue growth from January to April 2026 exceeded 100% year-over-year. The product will integrate large AI models and multimodal interaction technologies to enable environmental data feedback and intelligent decision-making. Taking the mid-to-high-end indoor delivery flagship product “Bella Pro” as an example, it has developed features combining retail scenarios with AI, including intelligent inventory monitoring, crowd characteristic recognition, AI-based dirt detection, and coordinated summoning.
Facing a trillion-dollar market, Pudu’s commercial cleaning robots have consistently led the industry in shipment volumes. Its core Magic Cleaning AI Agent system not only accurately identifies special surfaces such as adhesive residues and carpets but also enables customized detection of unique obstacles like cotton and plastic bags. Features derived from the system—including “garbage/stain heat maps,” “ultra-fast cleaning,” and “intelligent re-cleaning”—significantly reduce users’ daily deployment and operational costs.
In the industrial delivery sector, sales of Pudu’s industrial delivery products grew by 100% year-over-year in the first quarter of 2026. The flagship intelligent mobile robot series, T300, surpassed 4,000 units shipped in its first year on the market, achieving the fastest growth rate in the industry. The lightweight intelligent mobile robot T150 also surpassed 1,000 units shipped within four months of its launch. The products have been adopted across more than ten specialized industries, including automotive manufacturing, electronics and semiconductors, consumer electronics, PCB manufacturing, and warehousing and logistics. In May of this year, Pudu officially announced its entry into the autonomous forklift market.
Zhang Tao has a clear understanding of the industry; he believes the global sector is currently in a transitional phase between the commercialization and widespread adoption of specialized robots and the laboratory-scale validation of general embodied intelligence. Pudu, which has already achieved profitability and secured a leading position in specialized robotics, has both the confidence and strategy to be among the first to move into general embodied intelligence.
Pudu's embodied intelligence product portfolio consists of three lines: First, the quadruped robot line, featuring acrobatic capabilities, all-terrain obstacle negotiation, autonomous inspection, and following functions, suitable for tasks such as power line inspection, mapping, guiding the visually impaired, and disinfection. Second, the humanoids robot line, equipped with proprietary joint modules and a three-finger dexterous hand, focused on complex dual-arm collaborative flexible operations in industrial production and commercial services. Third, the humanoid robot line, which deeply enhances two core capabilities—“body structure” and “biomimetic cerebellum”—to accelerate integration into the deeper realms of human production and daily life.
Currently, Purdue Robotics' products are widely deployed across 16 scenarios, including retail, hospitality, industrial, warehousing and logistics, food and beverage, property cleaning, healthcare, entertainment and sports, education, public transportation, and services. Shipping volume and revenue continue to grow, with cumulative shipments exceeding 130,000 units and operations spanning 85 countries and regions worldwide.
According to the July 15, 2025 update of the Independent Market Research Report on Global Embodied Intelligence and Commercial Service Robots by the international consulting firm Frost & Sullivan, Pudu Robotics ranks first globally in commercial service robot revenue, shipment volume, and overseas expansion, with market shares of 25%, 23%, and 44% respectively. In the specialized commercial cleaning segment, Pudu Robotics holds a 29% market share by revenue, ranking first. Overall, Pudu is one of the few companies in the industry to achieve mass production across all product categories, global market penetration, and a clear path to profitability.
A long-term, multi-round博弈 hinges on using the depth of your product lineup to earn customers’ long-term trust through the prolonged and brutal market consolidation. Each expansion into new categories is made possible by leveraging the technological stack, contextual understanding, and channel resources accumulated through these battles—extending reusability to a broader range of robotic forms and achieving the leap from single-point breakthroughs to platform-level positioning.
Pudu's four product lines have established a robust, comprehensive defense that not only mitigates the risk of volatility in any single market but also creates a self-sustaining business ecosystem with global reach.
Behind the rollout of the "one brain, multiple forms" concept and the world's largest accumulation of real-world data across its entire product line is a robust technical foundation. "Technology and business walk on two legs; without either, you can't go far," said Zhang Tao.
In the era of embodied intelligence, where technology and scenario understanding are most rigorously tested, Pudu has its own strategic roadmap and core formula:
Physical Agent = Body + System + Skills
The embodied intelligence industry is evolving from “single-purpose execution tools” to “general-purpose physical agents.” The “one brain, multiple forms” architecture mirrors the ecosystem闭环 of the mobile internet era—“smartphones + operating systems + apps.” Pudu aims to rebuild this grand vision in the embodied intelligence industry: a unified intelligent brain that can adapt to various configurations, including specialized, humanoid, and human-like forms, significantly lowering development barriers and improving adaptation efficiency.
For a long time, embodied intelligence has faced three major bottlenecks in achieving practical application: first, architectural fragmentation leads to poor coordination—navigation handles movement while manipulation handles action, and the lack of a unified central system causes frequent breakdowns in complex tasks. Second, robots lack physical intuition—they can see objects but do not understand common-sense principles like gravity and friction, making fine manipulation akin to a blind man touching an elephant. Third, robots with different configurations are trained with separate, specialized models whose capabilities cannot be transferred across form factors; the deeper crisis lies in the fact that vast amounts of real-world data generated across different scenarios are isolated from one another and cannot be combined effectively.
To this end, Purdue has developed PuduFM, an embodied intelligence foundation model featuring three core technological dimensions: deep perception and reasoning of 3D space, future-oriented physical state prediction, and a learning mechanism that evolves through real-world interaction. By leveraging an understanding of the physical world, PuduFM enables unified operation of heterogeneous robots, bridging the gap from “execution” to “cognition.” In simple terms, PuduFM aims to help robots understand the physical world.
On the other hand, for robots to achieve scalable deployment, they must provide a unified physical agent system that enables developers to flexibly assemble existing capabilities like modular components, rapidly building applications without repeatedly reinventing the wheel. To this end, Pudu has developed PuduAgent—a general-purpose embodied intelligence platform for the physical world. If PuduFM is responsible for helping robots understand “how to get things done,” then PuduAgent acts more like the robot’s “task brain.” It can further comprehend user task objectives, autonomously break down complex workflows, dynamically invoke diverse capability modules such as navigation, manipulation, and perception, and coordinate multiple robots to accomplish long-range, complex tasks spanning multiple areas and processes.
Built on this unified underlying model, different types of robots act as distinct avatars sharing the same “intelligent brain”: greeting robots understand appropriate smiles and subtle gestures, delivery robots autonomously plan optimal routes for precise drop-offs, and cleaning robots dynamically adjust their strategies in real time based on floor materials and environmental changes.
This means that Purdue’s technological expertise accumulated in delivery scenarios can be efficiently transferred to entirely new fields such as cleaning and industrial applications, enabling rapid product development and consistent user experiences with only minimal targeted training—fundamentally transforming the R&D paradigm and deployment efficiency of service robots.
A multi-modal brain constructs an intricate framework, and data is the blood that gives muscles to this framework. The scale and quality of data determine the upper limit of embodied intelligence models' capabilities, while the efficiency of data acquisition affects the speed of model iteration.
However, an unspoken pain point for the industry is the lack of real-world data. Simulation data has limitations, while real-world data is prohibitively expensive. Tesla can gather massive amounts of real-world driving data from its hundreds of thousands of on-road vehicles, but in the field of embodied intelligence, no company can simply deploy hundreds of thousands of robots into the real world.
This is precisely what constitutes Pudu’s unreplicable advantage. As a leading embodied AI company with one of the largest global datasets of robot navigation, Pudu leverages its fleet of 1.3 million commercial robots deployed worldwide, generating 50 million hours of real-world data annually—including 36.5 million hours of authentic, effective, and diverse navigation data.
At the granular level of "operational data," industry-standard remote operation data is expensive and inefficient. Manual data collection, on the other hand, is prone to human motion distortion due to configuration differences. To address this, Pudu has developed a seamless data collection solution.
Pudu collaborates with global channel partners and customers to deploy seamless data collection devices, enabling operators to gather data without altering their daily work routines, achieving up to 1,580 hours of data collection per person annually. By rapidly aggregating 1,000 ecosystem partners, this creates an annual flood of 15.8 million hours of operational data.
This “scenario as collection, action as data” model not only reduces data collection costs by an order of magnitude but also ensures the physical authenticity and natural motion of the data. By leveraging vast amounts of seamless collection data, PuduFM can rapidly learn extensive prior knowledge and operational concepts. Building on this, reinforcement learning fine-tuning using correction and failure data enables large-scale, online, distributed policy updates and continuous evolution across robotic systems.
A brain-multi-form technical architecture, combined with a real-world data flywheel built from over 130,000 physical devices deployed globally, enables Pudu to transfer its accumulated understanding of real-world scenarios and data across industries to embodied general robots, achieving cross-product-line technology reuse for specialized, humanoid, and human-form robots.
The launch of the 03 PUDU D7 marks a new leap in humanoid robotics, and while delivery scenarios test a robot’s ability to navigate and plan paths in three-dimensional space, more complex real-world operational environments further challenge its understanding and manipulation of the physical world—not only must the robot “get there,” but it must also “grasp accurately, place steadily, and perform effectively.” This requires a leap from “legs” to “hand-brain coordination.”
Currently, embodied intelligent robots are rapidly entering real-world operational scenarios such as industrial manufacturing, warehousing and logistics, and retail supermarkets, but the industry still faces diverse explorations in technological approaches.
Traditional industrial robotic arm manufacturers are attempting to enhance their flexibility by integrating large models, but are limited by fixed workspaces and lack autonomous mobility and environmental awareness. Meanwhile, humanoid robot startups aim to cover a broader range of scenarios with a general-purpose design, yet the stability, maintenance costs, and maturity of the supply chain for bipedal structures under prolonged, high-frequency operations still require further validation.
This year at WAIC, the PUDU D7 offered another practical approach to embodied intelligence.
As a next-generation humanoid intelligent job partner designed by Pudu for real-world operational scenarios, the PUDU D7 adopts a technical approach combining an omnidirectional mobile chassis with flexible dual arms, integrating the precision manipulation capabilities of the arms with the stability and high efficiency of the mobile platform to achieve greater operational reliability and scenario adaptability while maintaining dexterous handling.
Currently, D7 supports a maximum load of 14 kg and a working height of 2 meters, enabling it to perform complex tasks such as material handling, shelf retrieval and placement, and internal transportation. It meets the high-frequency operational demands of industrial manufacturing and can gradually expand into additional real-world commercial environments requiring mobile manipulation, including warehousing, logistics, retail, and supermarkets.
For high-precision tasks such as dispensing and precision assembly, the D7 is equipped with a high-accuracy haptic sensor that enables real-time perception of contact force and pressure feedback, achieving millimeter-level force control. The system dynamically adjusts force application based on different materials and operating conditions, ensuring stable performance in scenarios like delicate picking, gentle pressing, and precise force application, thereby preventing damage to workpieces due to improper force. Additionally, the device delivers consistent thrust output, enabling autonomous execution of tasks such as cart placement and cargo handling, further enhancing operational efficiency in warehousing, logistics, and production environments.
More importantly, D7 is not just a product but also a significant application of Pudu’s “One Brain, Multiple Forms” technology framework in real-world operational scenarios. During tasks such as transporting, grasping, and placing objects, the robot continuously collects multidimensional data—including environmental perception, action execution, and human-robot interaction—creating a closed-loop data system spanning the entire task workflow. The PuduFM embodied intelligence large model continuously learns spatial relationships, operational logic, and task processes from these real-world data, steadily enhancing the robot’s decision-making and execution capabilities across diverse scenarios.
This means that, leveraging PuduFM and the unified intelligent foundation, D7 can continuously understand tasks, accumulate experience, and optimize action strategies in real-world environments, integrating the entire sequence of movements—such as moving, approaching, grasping, transporting, and placing—into a single intelligent system to enable continuous reuse of capabilities across tasks and scenarios.
For Purdue, industrial manufacturing is merely an important starting point for the practical application of embodied intelligence, not the end goal. As the unified intelligent foundation continues to evolve, this capability will extend further into additional real-world commercial scenarios such as warehousing and logistics, retail, and supermarkets. Through the "one brain, multiple forms" architecture, it will continuously migrate to more robotic forms, driving robots from performing single tasks toward becoming general-purpose intelligent assistants capable of adapting to multiple scenarios and tasks.
Leveraging its commercialization capabilities accumulated over the past decade in the business service robotics sector and its ongoing technological investments in embodied intelligence, Pudu Robotics announced in April this year the completion of a new round of financing nearing RMB 1 billion, co-led by Longgang Financial Holdings and Asia Investment Capital, with participation from BAIC Capital, Lanxi Technology, Shanghai Honghui, and multiple government-guided funds from the Pearl River Delta and Yangtze River Delta regions, along with renowned hard-tech investment institutions. Following this financing round, the company’s valuation surpassed RMB 10 billion.
For Pudu, this funding round signifies not only capital recognition but also the ability to continue investing in core research and development of embodied intelligence and global commercialization. As Zhang Tao stated, “The value of embodied intelligence lies not only in replacing repetitive labor but also in enabling robots to truly integrate into real-world production and daily life.”
This is also the path Pudu is most familiar with and excels in: developing the MT1 dry cleaning robot to address major cleaning challenges in warehousing, logistics, industrial workshops, and transportation hubs. Since its launch, it has become one of the world’s leading commercial cleaning robots. The T series of industrial AMRs, iterated to meet demands for heavy-load capacity and scene adaptability, has redefined the traditional model of expensive, customized industrial robots. In every specific, real-world scenario, Pudu dives deep and emerges with improved products—underpinned by the core belief that robots must grow from the front lines of actual industry.
As it extends into embodied intelligence, Pudu’s decade-long understanding of real-world scenarios and its accumulated data across 16 industries—including retail, hospitality, industrial, warehousing and logistics, food service, and property cleaning—constitute a formidable barrier. An industry淘汰赛 is approaching, and only those companies that truly understand the nuances of the physical world, possess authentic scenario data, and have the global commercial resilience to sustain growth will emerge victorious in the long run.