The robot standing over a white sheet of paper at Messe Berlin is not performing a trick. The brushstroke it is laying down—ink flowing with measured pressure, the tip widening as the stroke gains weight before lifting cleanly for the character's final element—is one of the most demanding contact-sensing tasks in robotics. The technology that makes it possible also allows every facility entered by AgiBot's A3 to be mapped and logged in precise detail. European buyers watching the calligraphy demonstration in Hall 25 this week need to consider both facts before signing a purchase order.

IFA Berlin 2026 opened this morning at Messe Berlin, and the calligraphy demonstration at AgiBot's Stand H25-371 in the IFA Next innovation zone has drawn the longest queues of the show's opening hours. That is hardly surprising: a humanoid robot executing one of humanity's oldest visual arts with the patience of a practiced calligrapher is exactly the kind of moment that stops a trade-show crowd. Less visible to the audience is the engineering behind the performance—and the legal framework that comes with hardware shipped from Shanghai.

How a Robot Writes Calligraphy—and What That Requires

Chinese calligraphy is considered China's supreme visual art form because it demands an extraordinary level of technical control. Every stroke requires the simultaneous modulation of three variables: downward force on the brush tip, which determines how widely the animal-hair tip spreads and therefore the width of the stroke; lateral angle, which controls the direction and flow of the ink; and movement speed, which controls ink density from start to finish. If any one of the three changes incorrectly mid-stroke, the character is ruined.

For a robot to perform this in real time, it needs what engineers call force transparency—the ability to sense force at the point of contact, at the actual brush-paper interface, without the signal being filtered or delayed by mechanical components between the finger and motor. Standard industrial robot hands use gearboxes to multiply torque. A gearbox allows a small motor to hold a large load, but it creates a mechanical separation between what the finger experiences and what the motor registers. A geared hand must infer contact force through position encoders and model assumptions. It cannot feel the brush dragging across paper as a human wrist does. Bhatia, Johnson and Mason's 2019 paper on direct-drive gripper force transparency documented this trade-off: high gear ratios deliver clamping force at the cost of the bidirectional force-motion signal needed for delicate manipulation.

AgiBot's OmniHand 3 Ultra-M—the dexterous hand system underpinning the A3's manipulation capability—eliminates that translation layer. Its 20 active degrees of freedom use fully direct-drive actuation, with no gearboxes or tendon transmission between motor and finger. The result is what the field calls mechanical transparency. In practice, for calligraphy, the A3's motors can detect and respond to the drag of the brush on paper during a stroke, adjusting force in real time as a human calligrapher would. Visuotactile sensors at all five fingertips and a tactile array in the palm provide a second data channel: miniature cameras inside the compliant fingertip structures observe surface deformation and translate it into contact location and force magnitude simultaneously.

The OmniHand 3 Ultra-M weighs 630 grams (1.39 lbs) and achieves ±0.2 mm positional repeatability. At IFA, it is paired with the A3 platform—a full-size humanoid standing 173 cm (5 ft 8 in) tall and weighing 55 kg (121 lbs). AgiBot says the platform adds an industry-first touch-sensing layer to its full-body control architecture, alongside a 360-degree multi-array microphone system and up to 12 kW of instant peak power. The touch-sensing layer extends real-time tactile awareness beyond the hands to the entire robot, allowing the A3 to respond to physical contact from any direction. That is relevant on the crowded trade-show floor, as well as in the factories and commercial environments where the robot is designed to operate.

This is the engineering that explains the calligraphy. It is also the engineering that makes the security framework below impossible to ignore.

Where AgiBot Stands in the Market

AgiBot—formally Zhiyuan Robotics, or AgiBot Innovation Shanghai Technology Co., Ltd.—was founded in February 2023 by Deng Taihua and Peng Zhihui, both former Huawei engineers. In November 2025, its A2 humanoid was recognised by Guinness World Records for walking more than 100 kilometres (62 miles) continuously. The company reached its 10,000th robot in March 2026 and its 15,000th in June 2026, with commercial deployments across more than 20 enterprises in factory, hospitality and service environments.

Independent market research firm Omdia ranked AgiBot first globally in humanoid robot shipments in 2025, attributing approximately 39% of global market share to the company. A subsequent analysis by Smart Analytics Global put AgiBot's first-half 2026 shipments at about 8,400 units, representing roughly 44% of global volume during that period.

The A3 made its European debut at AgiBot's UK Partner Conference in late June and early July 2026, which formally introduced the robot to European partners. William Shi, AgiBot's president for Europe and the Americas, described the UK as "a strategically important market for AGIBOT's global expansion." IFA is the A3's first major international trade-show platform. Shi is also speaking on the IFA programme stage today.

The strategic context is clear. The US Federal Communications Commission added foreign-made advanced robotic devices, including humanoid robots, to its national-security Covered List on July 28, 2026. The move effectively bars new AgiBot models from receiving the FCC authorisation required for import, marketing or sale in the United States. It closed the world's largest technology market to new Chinese humanoid models just six weeks before IFA opened. Europe has no equivalent restriction. For AgiBot, IFA is now the most commercially consequential week of the year.

IFA Next and the Chinese Robotics Surge

IFA Next, the show's dedicated innovation zone, has grown to approximately 300 exhibitors this year. Of roughly 1,900 exhibitors at IFA 2026 overall, approximately 932 are Chinese companies—close to half the total and a record for the show. Chinese manufacturers dominate the robotics section of that roster: Unitree, AgiBot, DEEP Robotics, EngineAI, Dobot, Cozio Robotics, Primebot and AiMOGA are all exhibiting in Hall 25.

IFA CEO Leif Lindner has described the moment as structural rather than temporary, pointing to China's manufacturing scale, investment levels and fully integrated supply chain as drivers of the pace of development.

The calligraphy demonstration reflects the show's biggest story: humanoid robots have become capable enough to perform delicate arts, while the country producing them at scale has also become the main commercial supplier for markets that the United States is restricting at home.

Robots on the Runway

AgiBot will take part in "Robots on the Runway" on the IFA Creator Stage live programme tomorrow, September 5, at 12:45 p.m. Berlin time (6:45 a.m. ET). It is the first event in IFA's history in which humanoid robots will walk a catwalk. Bipedal machines will demonstrate sensor-based navigation, motor control and movement capability in a live format. Other participants include Primebot Robots, EngineAI, Cozio Robotics, Dobot Robot, DEEP Robotics, Unitree, ZBy–Mind with Heart, Ollobot, Astrall Dynamics and AiMOGA, followed by a meet-and-greet.

The event will run simultaneously with a session by veteran technology analyst Sascha Pallenberg on the IFA Next Dream Stage, titled "The Humanoid Hype: Substance or Spectacle?" The session reflects IFA's recognition that scepticism and excitement about the category coexist in equal measure.

What European Buyers Need to Know Before the Purchase Order

AgiBot's presence at IFA is designed to support commercial engagement. Its Robot-as-a-Service model, launched at MWC 2026 earlier this year, offers the A3 at EUR 899 per day across 17 countries, with full technical support included. The model is aimed at events, exhibitions and short-term commercial deployments—the very buyers visiting IFA Next this week.

Before any buyer in Europe or elsewhere acts on that price, four dimensions of the decision require equal consideration. AgiBot's performance advantages are real; so are the gaps and legal obligations that accompany them.

Performance and Generalisation Gaps

AgiBot's deployment data from Longcheer Technology's tablet manufacturing line in Nanchang—99.99% task success across 64,800 production cycles—is company-reported and has not been independently audited. More revealing when evaluating underlying capability, AgiBot's own GO-1 foundation model improved from 46% to 78% task completion in real-world scenarios after training on more than one million robot trajectories. The 78% figure represents meaningful progress, but remains below the 95% or higher reliability threshold industrial operators typically require before deploying robots in unsupervised production environments. A Morgan Stanley buyer survey found that only 23% of prospective industrial customers were satisfied with currently available humanoid robot products.

Independent engineering analysis has also found that US firms including Physical Intelligence, Skild AI and Google's Gemini Robotics still lead in generalisation capability—the ability to handle tasks for which a robot was not specifically trained—while Chinese providers including AgiBot currently lead in cost integration for defined workflows. Those advantages are not the same, and both matter before a robot can be treated as general-purpose capital equipment.

Benchmark Reliability

AgiBot's published performance figures, including the Longcheer deployment metrics and GO-1 foundation-model benchmarks, come from the company's own evaluations. As of this writing, most of the claims have not been independently validated by third parties. This is standard for the category in 2026: nearly all Chinese humanoid robot producers are in a similar position, while independent auditing infrastructure for embodied AI remains at an early stage. Buyers should treat unaudited company benchmarks with appropriate caution, particularly when considering sustained commercial deployment rather than event or exhibition use.

Ecosystem and Operational Considerations

The A2 Ultra holds four major international certifications: CR, CE-MD, CE-RED and FCC. The CE-MD, or EU Machinery Directive, and CE-RED, or EU Radio Equipment Directive, certifications indicate that AgiBot has met European safety and radio-emissions standards. They address hardware safety and radio compliance, but do not alter the Chinese government's legal authority over AgiBot as a Chinese company.

AgiBot's RaaS model and European partner network reduce some operational friction for buyers who do not want to manage robot maintenance directly. Documentation and support infrastructure for international operators are still developing. Buyers deploying AgiBot hardware in complex or regulated environments should budget engineering time for integration beyond the base equipment cost.

China's National Intelligence Law: What It Means for the A3's Sensor Suite

This is the dimension that no hardware specification addresses.

AgiBot is headquartered in Shanghai and operates as a Chinese company. Article 7 of China's National Intelligence Law of 2017 requires all Chinese organisations and citizens to "support, assist, and cooperate with state intelligence work in accordance with the law." The obligation is not conditional. It does not change with the company's stated privacy policy, the physical location of its servers, the jurisdiction in which its robots are deployed or whether the company has European subsidiaries. It is the operative law of the company's home country and applies regardless. Research by the Carnegie Endowment on Chinese data-access obligations says this structural compulsion extends to connected devices and the data they collect.

China's Cybersecurity Law, enacted in 2017 and amended effective January 1, 2026, explicitly extends government technical-support obligations to AI system operators. Its Data Security Law of 2021 adds data-localisation requirements and requires companies to report newly discovered security flaws to Chinese authorities within 48 hours, before patching or disclosing them to customers.

The A3's sensor suite makes this disclosure materially important. Its full-body architecture includes continuous visual data from RGB-D cameras and fisheye lenses; LiDAR spatial mapping equivalent to a detailed architectural scan of any facility where the robot operates; centimetre-level positioning through GPS, RTK and UWB fusion; audio capture through its 360-degree multi-array microphone system; and contact-force data at fingertip and palm level through the OmniHand's visuotactile sensors. An A3 operating on a factory floor for several weeks would, incidentally, build a precise map of the facility and a detailed record of its production processes. No published independent security audit of AgiBot's firmware or data-transmission behaviour has been released.

This is a structural gap distinct from Unitree Robotics, where researchers documented specific Bluetooth vulnerabilities and hidden telemetry paths in 2025. For AgiBot, the risk is legal and structural rather than a documented technical backdoor. The absence of an audit is itself a material fact for any procurement decision.

Practical measures available to buyers include network segmentation, such as air-gapping the robot's wireless connectivity from core production networks; disabling unnecessary connectivity during sensitive operations; and conducting a legal review before deployment in environments subject to export controls, ITAR obligations or classified operations. None of these steps changes the structural obligation under Article 7.

Abel Deng, AgiBot's president for the Middle East and Asia Pacific, has described 2026 as a critical milestone as the company approaches scaled real-world deployment across factories, commercial facilities and everyday operational environments. The milestone is real, and so is the legal framework. Both belong in any deployment decision.

The Full Decision Framework

AgiBot at IFA 2026 offers four things simultaneously: a genuine mechanical achievement—a 55 kg (121 lb) humanoid that writes calligraphy through real-time contact sensing; a production-scale deployment record of 15,000 units across more than 20 commercial enterprises; a commercially accessible entry point of EUR 899, or approximately US$1,043, per day through RaaS with no long-term commitment required; and a fixed legal obligation to Chinese state intelligence that applies to every unit regardless of where it operates.

European buyers at IFA face a regulatory environment that the United States has already resolved in one direction. The FCC's July 28, 2026 Covered List action closed the US market to new AgiBot models. Europe has made no equivalent determination. The adoption decision in Europe therefore rests fully with individual buyers—and must be made with all four dimensions of the framework in view, not only the one most visible from across the trade-show floor.

Exchange rate as of September 4, 2026; conversions are approximate.

Frequently Asked Questions

What makes AgiBot's A3 humanoid capable of writing calligraphy—and why does it matter beyond the performance?

Chinese calligraphy requires simultaneous, real-time control of brush pressure, angle and speed, making it one of the most demanding contact-sensing tasks in robotics. The A3 achieves this through the OmniHand 3 Ultra-M's fully direct-drive actuation, which removes gearboxes between motor and finger and provides what engineers call force transparency: the ability to sense the brush-paper interface directly at motor level, without mechanical filtering. Visuotactile sensors at all five fingertips capture contact location and force magnitude simultaneously. The same architecture that makes calligraphy possible also makes the robot a highly capable environmental sensor, which is critical to understanding the legal framework described above.

What does China's National Intelligence Law actually require of AgiBot, and does it apply in Europe?

Article 7 of China's National Intelligence Law of 2017 requires all Chinese organisations to support, assist and cooperate with Chinese state intelligence work on demand. The obligation applies to AgiBot as a Shanghai-headquartered company regardless of where its robots are deployed, what its privacy policy says or where its servers are located. Europe has no equivalent to the US FCC Covered List restricting Chinese robots, so European buyers face this legal condition without a regulatory framework that has assessed it on their behalf. The A3's cameras, LiDAR, microphones and force sensors collect facility-level operational data that this obligation could legally compel AgiBot to share with Chinese intelligence agencies, according to Carnegie Endowment analysis of China's data-access framework.

How reliable are AgiBot robots in real-world commercial settings, beyond trade-show demonstrations?

AgiBot's most detailed deployment data comes from a Longcheer Technology tablet manufacturing line in Nanchang, China, where company-reported data claim a 99.99% task-success rate across 64,800 production cycles. These figures have not been independently audited. AgiBot's GO-1 AI foundation model achieved 78% task completion in real-world scenarios after training on more than one million trajectories—genuine progress, but below the 95% or higher threshold typically required for unsupervised industrial deployment. A Morgan Stanley buyer survey found that 23% of prospective customers were satisfied with currently available humanoid robots generally. The technology is advancing quickly; the gap between demonstration performance and sustained production reliability remains the honest measure of its current position.

Is AgiBot's Robot-as-a-Service pricing a safe way to evaluate the technology without a full purchase commitment?

The RaaS model, starting at EUR 899 per day for a minimum one-day rental across 17 countries, significantly lowers the financial barrier to evaluation. The legal framework, however, does not change with the contract structure. A rented AgiBot robot remains an AgiBot robot subject to Chinese law throughout the rental. Buyers evaluating the technology through RaaS should still use network segmentation, avoid deployment in environments with ITAR or classified-data obligations, and treat the rental period as an operational trial of the hardware rather than a way to avoid the legal question.

Originally published on Tech Times