Embodied Intelligence 2026: Farewell to Empty Storytelling, Practical Implementation Reigns Supreme

The World Robot Conference held in Yizhuang, Beijing is growing increasingly popular year by year.

Judging from the on-site situation, UBTECH, known as the "first stock of humanoid robots", still stays at the forefront of the industry. This year, it brought demonstrations of application scenarios covering industry, commercial services and household consumption. The commercial service humanoid robot Walker C1 provides reception, hosting and explanation, performance interaction and teaching services. The ultra-realistic humanoid robot U1 delivers emotional value. Nearly 10 industrial humanoid robots including Cruzr Y1 and Cruzr S2 form a miniature robot factory, which completes full real-scene operations such as box handling, palletizing and depalletizing, loading and unloading of auto parts, and sorting of small express parcels...

Further research shows that behind this is the full-stack self-developed technology system built by UBTECH, which consists of "foundation model — world model — action model", with one set of embodied intelligence foundation supporting the three scenarios at the same time.

Coupled with a large number of landing cases and cooperation cases highlighted at the booth, this is one of the very few booths at this year's World Robot Conference that can fully present a complete closed loop of technology, products, scenarios and delivery.

"Previously, people might prefer to make stories more attractive and appealing. But from this year onwards, we have to focus on actual orders, deliveries and data," JIAO Jichao, Vice President of UBTECH and Dean of Embodied Intelligence and Humanoid Robotics Research Institute, told 36Kr.

In 2026, the first year of embodied intelligence's marketization, UBTECH, a 14-year-old enterprise with marketized orders and full-stack technologies, is also one of the few players that can translate "technological leadership" into "industrial leadership" at the earliest stage.

Industry, Commercial Services, Household Consumption: Three Scenarios, Progressing Step by Step and Interconnected Closely

2026 is recognized as the first year for the implementation of embodied intelligence, a consensus has been almost reached across the whole industry. The biggest bottleneck of the current industry is shifting from "whether we have the ontology and model" to "whether we have landing scenarios with sufficiently high value".

Specifically for UBTECH, it has placed its bets on the three scenarios: industry, commercial services and household consumption.

The first is the industrial scenario. Industries such as manufacturing, automotive, 3C and logistics are highly standardized with clear processes and repetitive tasks. The value created by robots can be directly calculated through efficiency, stability, labor cost and return on investment, which makes it the easiest starting point for the embodied intelligence industry to form a commercial closed loop at the current stage.

Walker S2 performs collaborative sorting in a real factory scenario

As early as around 2024, UBTECH has started to deploy its self-developed Walker S series humanoid robots in factories, covering many industrial fields including automobile manufacturing, 3C manufacturing, smart logistics, semiconductor manufacturing, aviation manufacturing and industrial data collection, which serves as the core cornerstone for the implementation of UBTECH's embodied intelligence technology.

According to UBTECH's 2025 annual performance report, its total annual revenue exceeded 2 billion yuan. The full-size embodied intelligent humanoid robots (non-remote-controlled and non-toy, with a height of over 160cm) mainly used in factories generated a revenue of 820 million yuan with sales volume of 1079 units, ranking first in the world, which is widely regarded by the market as a key node for embodied intelligence to move from "proof of concept" to "real implementation".

The second is commercial services. Factories require robots to "do things correctly", while commercial scenarios require robots to further learn to "understand people, serve people, and interact and collaborate with people".

In this context, the newly released full-size commercial service humanoid robot Walker C1 by UBTECH is equipped with capabilities including navigation, inquiry, language understanding, whole-body coordination and multi-modal interaction, which enables humanoid robots to have core capabilities for implementation in fields such as supermarkets and malls, exhibition halls, scientific research and basic education.

In addition, UBTECH also premiered the Walker C1 EDU Tutor for basic education scenarios at this World Robot Conference. Equipped with the UIRO intelligent teaching assistant system, it can independently complete teaching tasks such as autonomous teaching, problem-shooting explanation, intelligent Q&A and multi-modal classroom interaction, targeting the intelligent upgrading of the basic education industry.

The household consumption scenario is the most special one, and the ultra-realistic humanoid robot is also UBTECH's most "viral" product this year.

In June this year, at UBTECH's annual global launch event, UBTECH officially unveiled the ultra-realistic humanoid robot U1, which includes three configurations: the half-body version U1 Lite, the high-end full-body version U1 Pro, and the high-dynamic full-body version U1 Ultra. U1 Lite and U1 Pro are exhibited at the WRC site, with selling prices of 119,800 yuan and 169,800 yuan respectively, while U1 Ultra is mainly designed for commercial scenarios.

This is almost the first time in the industry that a well-known robot enterprise breaks away from the idea of focusing on "practicality" and "low-price involution", takes the initiative to enter the emotional companionship track, and launches a consumer-grade humanoid robot product that users can actually purchase.

This series of ultra-realistic humanoid robots is equipped with a 33-axis facial tendon network, which can realize more than 30 composite micro-expressions with an emotion recognition accuracy of over 90%. It is also equipped with UBTECH's Resonance-LM emotional large model, which can identify more than 20 fine-grained emotions. Instead of focusing on tool attributes, it truly aims to let robots enter families and daily lives.

The three scenarios of industry, commercial services and household consumption progress step by step and are closely interconnected.

The industrial scenario not only has the clearest demand and highest implementation certainty, but also serves as an indispensable technical reserve for robots to construct the physical world and understand the real world, which is equivalent to obtaining customers, revenue and a real-world entrance that continuously generates training data at the same time, and then gradually spills over the software and hardware capabilities such as ontology, supply chain and embodied intelligence model to commercial and household scenarios.

The three scenarios combined have pre-emptively occupied all three major markets that humanoid robots may enter. What makes UBTECH rare is that it not only occupies the industrial scenario that generates revenue first at present, but also makes early bets on the two larger long-term markets of commercial services and household consumption, with a rare cross-cycle scenario portfolio.

Full-stack Self-developed Embodied Intelligence Technology Enables Full-scenario Implementation of Humanoid Robots

"Many companies developing embodied intelligence large models often build a ready-made ontology to run, but this cannot maximize the hardware efficiency. As an ontology manufacturer, we define capabilities from the very beginning of product design, and then combine algorithms to achieve deep integration of software and hardware," JIAO Jichao told 36Kr.

He gave an example that without deep adaptation of software and hardware, when an ordinary embodied intelligence large model "runs", the hardware efficiency can only reach 60%-70%, which is undoubtedly a huge waste in the era of scarce computing power.

This is also why, in the field of basic technology, UBTECH has chosen the most difficult path — full-stack technology.

At present, UBTECH has built a complete full-stack robot technology system covering servo drives, joints, motion control, ontology control, operating systems, world models, and "brain/cerebellum". By the end of 2025, UBTECH owned 2985 authorized patents, including 1742 invention patents and 508 overseas authorized patents, and it has also led the formulation of 8 national standards for humanoid robots.

Especially in the currently most concerned field of embodied intelligence large models, UBTECH has built a complete model system of "foundation model — world model — action model". Its self-developed foundation large model Thinker breaks through the industry's technical bottlenecks with "small parameters, high performance and full open source"; its Thinker-WM world model ranks first in the authoritative embodied intelligence evaluation benchmark Libero, and the Thinker-VLA model can increase the reasoning efficiency by 176% in industrial scenarios, endowing robots with multi-dimensional interaction capabilities such as dialogue, inquiry, expression, eye contact and memory in commercial and household environments.

The current embodied intelligence industry is gradually shifting from competition of "single-point capability" to competition of "system capability".

Unlike large language models that highly depend on software performance, for robots in the real physical world, any short board in the brain, body, motion control, sensors, computing power, battery, data and manufacturing may directly lead to task failure.

More than ten years of accumulation in robot system technology makes UBTECH significantly different from most current embodied intelligence and robot manufacturers. It not only has a powerful motion control system for joints, ontology and motion control, but also conducts cutting-edge research on VLA, Foundation Model and world model, and even carries out cutting-edge exploration such as the U1 Resonanc-LM emotional large model, owning the most comprehensive and system-level "top to bottom" technology stack.

The real value of such powerful full-stack capability is to regard robots as a systematic project, continuously optimize and improve system efficiency, obtain continuous feedback between technology and scenarios, form a technology flywheel, and finally realize scenario implementation and enter daily lives.

Embodied Intelligence 2026: Farewell to Stories, Implementation is King

In the past few years, the embodied intelligence industry has never lacked stories.

Capital, resources, entrepreneurs and technical talents have all poured into the industry rapidly. The performance of ontologies has been continuously refreshed, and breakthroughs in technologies such as VLA and world model have further raised the market's expectations for general-purpose robots.

However, all technology industries will eventually go through the same process: shifting from "whether the technology is feasible" to "whether the business is feasible".

In 2026, embodied intelligence is reaching this node.

At this year's World Robot Conference, the industry trend has completely shifted. The dance, somersault and stunt performances that can be seen everywhere in pavilions in previous years have been greatly reduced, and "getting to work" has become the main theme. Most manufacturers no longer prioritize showing off skills, but directly display application scenarios and compete for the autonomous continuous operation capabilities of robots.

To achieve production-line level continuous operation, the requirements of humanoid robots for hardware and computing power are completely at a different dimension from those of performance robots.

The "dancing robots" focusing on dynamic performance are mostly about 1.2 meters in height, with lightweight design; their joints use planetary reducers to prioritize instantaneous explosive force; they are equipped with low-computing-power chips to run motion control models, which only need to execute pre-arranged motion trajectories. The whole set of movements is pre-programmed for playback, without the brain's real-time understanding of the environment and autonomous handling of unexpected situations.

As for the "working robots" for industrial scenarios, they are about 1.7 meters in height, with multiple sets of visual sensors and force sensors mounted on the fuselage, as well as end effectors such as dexterous hands and grippers; the upper limb joints use harmonic reducers with high positioning accuracy, which supports long-term continuous load of robots and high-frequency force interaction with the environment; the ontology brain adopts NVIDIA Thor high-computing-power chips of over 250T to run embodied intelligence large models locally, which can independently complete environment perception, task decomposition, dynamic strategy adjustment and retry after operation failure, truly realizing autonomous work in the real physical world.

Different scenarios lead to great differences in products from hardware ontology, cerebellum to brain, which cannot be directly reused.

UBTECH's choice is very clear: not to pursue short-term amazing performance on the stage, but to focus on application scenarios that can be truly implemented, continuously reduce the ontology cost through mass manufacturing. At the same time, when robots enter real environments such as factories, commercial places and households, they can continuously generate real-world data to train models in reverse, continuously improve robot capabilities and spill over to other scenarios.

The truly valuable flywheel of an embodied intelligence company can start to rotate in this way.

While the industry is generally still looking for the Product-Market Fit (PMF) of humanoid robots and debating when they will be applied on a large scale, UBTECH has taken the lead in operating in the three major application scenarios, starting to solve the problems of production capacity, delivery quality and large-scale replication, and targeting the competition at the next dimension.

This is also the key to understanding why UBTECH enters the industrial, commercial and household markets at the same time; what it is looking for are the three key variables of embodied intelligence in markets of different maturity levels: revenue today, data tomorrow, and general capabilities the day after tomorrow.

There is no doubt that 2026 will be a major watershed for the embodied intelligence industry.

When the industry moves from the technology-intensive period to the stage where technology, manufacturing and business are delivered at the same time, the one who can take the lead in translating "technological leadership" into "industrial leadership" will have the opportunity to lead the industry pattern in the next stage.