A comprehensive analytical review of the "Humanoid Robots and Embodied Intelligence Going Global Forum. Greater Bay Area Station," organized by TΓV Rheinland. Nine presentations, 18 analytical sections, key conclusions on commercialization, technical barriers, global regulation, and international market entry strategy.
The TΓV Rheinland forum in the Greater Bay Area (GBA) became a landmark event of 2026, bringing together key players in the humanoid robot and embodied intelligence industry. Nine presentations covered the full spectrum of challenges: from technological bottlenecks (motors, dexterous hands, sensors) to the multi-layered regulatory landscape of the EU, North America, UK, and Asia-Pacific. The forum's key conclusion: safety and compliance are not a barrier, but a passport to global markets.
The forum was opened by James (host, TΓV Rheinland), who introduced the expert team: Li Yonggang (General Manager), Li Jie (Certification Expert), Chen Wenhua (Regional Director), and others. In his opening remarks, Li Yonggang emphasized: TΓV Rheinland is an organization with more than 150 years of history, of which 40 years in China, and it is ready to accompany Chinese robot manufacturers on their path to global expansion.
China's 14th Five-Year Plan achieved two target metrics: 50% domestic robots in the internal market and 50% domestic key components. According to IFR 2024 data, China has remained the global leader since 2014. Japan's share of the global market has fallen from ~40% to ~20%. The service robot market (10.34 million units) exceeds the industrial market (300β400K) by a wide margin.
The forum program covered: a commercialization roundtable (moderator Li Guochang, participants: Tan Weijia from the Shenzhen Robotics Association, Yu Zong. CEO Kensing Technology, Lei Yang. Tonglian Dynamics, Wang Zong. Tuobang Electric, Li Dong. TΓV Rheinland); performance testing methodology (Liu Lirong); GBA trends (Dr. Tan Weijia); two presentations on the EU AI Act (Tang Yilin); cybersecurity (Fan Zong and Liang Liubo/Byron); North American compliance (Li Shilin); functional safety (Li Qin).
"We've heard many grandiose predictions and seen capital madness. How many of these promises have we actually eaten? How many companies are truly earning, and how many are living on investments?"
β Li Guochang, roundtable moderator, TΓV Rheinland
The roundtable opened with the forum's sharpest question: when will humanoid robots start generating real profit? Li Guochang noted the paradox: on one hand, media hype (appearances on CCTV, investment rounds), on the other, most companies have not yet reached self-sustainability.
Lei Yang (Tonglian Dynamics) soberly stated: "At this moment, commercialization has not yet arrived, we are at the validation stage. Robots are being purchased by research institutes and universities for secondary development. Real commercialization will begin next year or the year after."
Tan Weijia (Shenzhen Robotics Association) offered a philosophical perspective: in his 13 years in the industry, he is constantly asked "when?" In his view, embodied intelligence is not something mystical, the first generation of embodied intelligence was AGVs: autonomous perception, autonomous decisions, autonomous execution. The transition is gradual, and the key question is not technical capability, but the customer's willingness to pay.
| Scenario | Status | Time Horizon | Key Drivers |
|---|---|---|---|
| Scientific Research / Universities | β Current primary market | Now | Secondary development, training |
| K-12 Education | π‘ Early commercialization | 2026+ | STEM, programming for children |
| Industrial Pilots | π‘ Pilot projects | 2027β2028 | Labor shortage in Japan/Korea |
| Home Assistants | π΄ Consumer demand exists, technical readiness absent | 2028+ | Cost reduction, versatility |
Yu Zong (Kensing Technology), whose exoskeletons already have commercial applications, compared the trajectory of AI robots to 3D printers and drones: "They went through 7β8 years of refinement before becoming mainstream. The front part of the journey is always longer than expected, the back part, shorter. We deeply feel: the problems are not in the algorithms, but in engineering, scenario adaptation, and service quality."
The second topic of the roundtable was technological limitations. Moderator Li Guochang framed the problem: "The robot's head is already advanced, but the limbs are lagging behind." The discussion revealed three critical areas:
Wang Zong (Tuobang Electric), specializing in transmission modules for dexterous hands ("η΅ε·§ζ"), compared Chinese developments to the leaders. Tesla and Figure. In his assessment, the main difference lies in the dexterous hands. "Many companies do beautiful demonstrations: they box, run, pose. But when it comes to actually grasping objects in production, small parts simply cannot be picked up. This is the biggest pain point."
"We bought several Chinese dexterous hands for testing. In demo mode, everything is perfect. But when you start actually assembling β holding with the left hand, placing with the right β it doesn't come together. Small objects simply cannot be gripped. Finger sensors are a huge problem. Whoever breaks through this β will win."
β Wang Zong, R&D Director, Tuobang Electric
Yu Zong added that motors for robots require a "reinvention" of architecture, just as the emergence of external rotors for drones changed the entire industry. Key metrics: power density, torque density, heat dissipation. When integrated into a compact body, the requirements are extreme. "China's advantage lies in engineering layering: just as chips went through Moore's Law, motors can go through engineering optimization."
A separate problem is tactile sensors on the fingertips. Wang Zong noted that this is an interdisciplinary challenge at the intersection of materials science and microelectronics. Without a breakthrough in this area, dexterous hands will remain a demonstration tool rather than a production one.
Yu Zong noted a non-obvious fact: "Most robots are still assembled by hand. You cannot order an automated line for a robot the same way you can for a car or a phone. This infrastructure has to be built from scratch."
The automation paradox: an industry meant to automate other manufacturing itself suffers from a lack of automated assembly lines. This lengthens the preparatory period before scaling.
Dr. Tan Weijia described the unique GBA supply chain path: Industrial Grade β Automotive Grade β Consumer Grade. This is the evolutionary ladder that robotics climbs as it moves from factory floors to consumer homes.
Industrial Grade: the ability to solve real production tasks. Automotive Grade: safety standards for human-machine interaction without protective barriers, critical experience from BYD and Chinese EV companies. Consumer Grade: GBA's unique advantage, a consumer electronics ecosystem with no parallel in the world.
According to Dr. Tan, the 15th Five-Year Plan (expected in June 2026) may account for the "Honor Effect": when consumer electronics manufacturers enter the robot market, their supply chain management competencies "destroy" traditional industrial approaches. The plan drafter from MIIT already called Honor's victory in the smartphone market a "landmark event" for robotics.
"You're shipping a robot to Europe. Who will insure it? If it causes harm in a factory β who pays? No standard β no insurance. And without insurance, no enterprise will buy your robot."
β Yu Zong, CEO, Kensing Technology
The roundtable participants identified five key barriers when entering international markets:
| # | Barrier | Description | Speaker |
|---|---|---|---|
| 1 | Insurance | The absence of standards makes underwriting impossible. The Shenzhen Association promotes an "insurance as CASCO" model β a fixed premium per unit of product. | Yu Zong, Tan Weijia |
| 2 | Localization | Instructions, labels, marketing materials β adaptation by native speakers is necessary. "Seemingly a trifle, but it can stop sales." | Yu Zong |
| 3 | GDPR & Data | Robots collect enormous volumes of personal data. GDPR fines (up to Β£10M or 4% of annual revenue) make compliance existential. | Lei Yang |
| 4 | Service Infrastructure | Consumers compare buying a robot to choosing a city based on healthcare quality. Without after-sales service, the product is unviable. | Lei Yang |
| 5 | Solutions, Not Hardware | It is not enough to sell the robot's "body." You need to offer ready-made use scenarios adapted to local conditions. | Lei Yang, Tan Weijia |
Tan Weijia added an important strategic tip: "Change your role, instead of being a solution provider, become an enabler. Sell modules, components, allow local partners to build the final solution. This reduces political risks and data problems."
The battery problem was discussed separately. Tan Weijia gave an example: under new EU rules, the battery importer bears responsibility for its full lifecycle, including disposal. With a battery service life of ~3 years and 5-year liability, costs double. Li Dong (TΓV Rheinland) emphasized the need for certification as a tool to reduce these risks.
The final topic of the roundtable was the Chinese phenomenon of "nΓ¨ijuΗn" (involution, internal race to exhaustion) in robotics. After the 2026 CCTV Gala, the market flooded with startups competing on specs and price.
"Competition is inevitable β it's a jungle. But the real question is: how to make it meaningful? Compete on quality, safety, service β not on price to the point of suicide."
β Lei Yang, Tonglian Dynamics
| Speaker | Position |
|---|---|
| Wang Zong | An urgent need for originality and genuine innovation. "Copying and lowering price is a low-intellect strategy that kills entire industries." |
| Lei Yang | A call for creating high industry standards as a barrier against dumping. "We need a functional safety standard, an operational precision standard, an energy consumption standard." |
| Yu Zong | Competition is inevitable, the one who looks further survives. "In any industry, even in the red ocean, someone is earning. That means they have something." |
| Li Dong | Compliance as a differentiator. "Don't lie on the path to market entry β ensure compliance first." |
| Tan Weijia | It's simply too early for competition now β automakers and internet giants have not yet entered seriously. "If you don't create a barrier now, when the big players arrive β you won't even have a chance to compete." |
Tan Weijia warned: "Automotive companies have not yet begun mass production of robots. Internet giants have not yet gone ALL-IN. Architecture convergence has not been achieved. If you don't create a technological barrier now, when the big players arrive, most simply won't have the ability to compete."
"Safety is the entry ticket. But safety β functionality. We've seen too many robots that passed certification but failed on the real factory floor."
β Liu Lirong, R&D Manager, TΓV Rheinland
Liu Lirong presented TΓV Rheinland's performance testing methodology, built around four dimensions:
| Dimension | What Is Assessed | Key Metrics |
|---|---|---|
| Mobility & Navigation | Movement in dynamic environments | Path efficiency, obstacle recovery, stuck percentage |
| Fine Manipulation | Grasping precision and adaptability | Positioning repeatability (Β±mm), force control smoothness, grasp success rate |
| Environmental Adaptability | Operation in extreme conditions | Illuminance range (0β100,000 lux), temperature, humidity, surface types |
| Reliability & Lifespan | Long-term stability | Accelerated testing, charge cycles, gear wear, impact resistance |
Key insight from Liu Lirong: "Many companies pass basic electrical and network certification and think they are export-ready. But on an old factory floor with oil stains or on a thick office carpet, robots fall, slip, lose orientation. The cost of a mistake in the export market is not a product return, but the loss of the entire market."
Special emphasis in fine manipulation testing: the robot must grasp objects of varying rigidity, from a cardboard box to a glass cup and a raw chicken egg. Grip force is measured to prevent crushing fragile items. This test reveals the gap between "beautiful demos" and real utility: "In the lab, a robot hand with 20 degrees of freedom poses flawlessly. But hand it an egg or a small bearing from a table, and you'll see the real level of technology."
Special emphasis was placed on visual and LiDAR perception: testing under extreme conditions, from complete darkness (2 lux) to bright direct light (1,500 lux), including mirror surfaces and glass walls that can "blind" the sensors.
TΓV Rheinland has built 8 test capability matrices, covering the complete robot evaluation cycle:
| # | Matrix | Content |
|---|---|---|
| 1 | Kinematics & Mechanics | 3D force platforms, foot pressure analysis, torque measurement, motion energy efficiency |
| 2 | Gait & Balance | Step phase breakdown, center of mass distribution in statics and dynamics, fall prevention algorithms |
| 3 | Algorithm Validation | Comparison of simulation with real motion capture data, limb and finger force control |
| 4 | Complex Environment Adaptability | Treadmill with variable speed/incline/load, stairs, uneven surfaces |
| 5 | Human-Machine Interaction | Interactive torque measurement, exoskeleton pressure curves, collaborative tests |
| 6 | Safety & Diagnostics | Comparison of expected and actual movements, deviation detection, peak pressures |
| 7 | Multi-Device Integration | Synchronization of optical motion capture + force platforms, unified timestamp |
| 8 | AI + Functional Modules | Visual perception tests, spatial perception (3D cameras), long-duration tasks, robustness |
Commercial value: TΓV Rheinland issues performance certificates for successful products. Examples: Kensing exoskeleton certification through human-subject experiments, "efficient mowing" certification for lawn-mowing robots, "anti-tangling protection" certification for robot vacuum cleaners. Cooperation with German magazines (Wissenstechnik, Diva) strengthens consumer trust.
Liu Lirong emphasized: "Safety certification is the entry ticket. Cybersecurity and AI Compliance are the protective moat. But performance testing determines the productivity ceiling, how far you can go in the market."
"China has been the global leader since 2014. Japan's global market share has fallen from roughly 40% to roughly 20%. Service robots β 10.34 million units. Industrial β 300β400 thousand. These are different universes."
β Dr. Tan Weijia, Secretary-General, Shenzhen Robotics Association, based on IFR data from April 2026
Dr. Tan Weijia presented exclusive data obtained from Dr. Susanna (IFR Secretary-General), who visited the Association's event in April 2026:
| Indicator | Value | Context |
|---|---|---|
| China β leader since | 2014 | Since the 14th Five-Year Plan (2016) β largest consumer; since 2018 β largest producer |
| Industrial Robots | 300β400K /year | Global market, IFR 2024 data |
| Service Robots | 10.34M | 30+ times larger than industrial |
| Two 50% (14th Plan) | Achieved | 50% domestic robots, 50% domestic components in the internal market |
| Japan's Share | ~40% β ~20% | Decline in recent years (unofficial data from Japanese colleagues) |
| Unstructured Environments Market | 3 times larger than industrial | Per Association classification: industrial vs non-industrial |
An interesting detail: Bambu Lab (ζη«Ή), a leading 3D printer manufacturer, 10% of all production is used for robot hardware components. This figure is growing and signals the emergence of a new DIY market in robotics, from professional developers to hobbyist makers.
AGV as the first generation of embodied intelligence. Dr. Tan demystified the term "embodied intelligence": "AGV IS embodied intelligence. Nothing mysterious: autonomous perception, autonomous decision-making, autonomous execution. AGV fully meets the definition. It evolves gradually, the problem is not technological impossibility, but customer readiness to accept the new." This is a key thesis: the transition to humanoid robots is evolutionary, not revolutionary. AGVs, collaborative arms, exoskeletons, and only then humanoids.
Dr. Tan introduced the concept of the "nurturable robot" (ε »ζεΌζΊε¨δΊΊ): a transition from the "developer" paradigm to the "maker" paradigm. "Foundation models consume too many societal resources. But if you give people a modular kit and the ability to train the robot themselves, the market will open up in a completely different way."
A deep analysis of the Greater Bay Area (GBA) competitive advantages, presented by Dr. Tan:
| Dimension | GBA Advantage | Global Context |
|---|---|---|
| Comprehensiveness | The entire chain: from microchips to finished robot within a 1-hour radius | No other region in the world has such concentration |
| Automotive Experience | BYD + new automakers: safety-critical engineering | Competency transfer to robotics is already underway |
| Consumer Electronics | Honor, DJI, mobile device ecosystem | Experience in low-cost mass production |
| AI Talent | Algorithmists in Beijing/Shanghai + hardware in Shenzhen | Horizon Robotics: model has not yet converged |
| Policy | "Jiebang Guashuai" of Nanshan District | 10M RMB demand β 5M RMB support (2.5M advance) |
The "Jiebang Guashuai" (ζζ¦ζεΈ ) policy of Nanshan District in Shenzhen deserves special attention. Dr. Tan, as the author of this policy, explained its essence: companies nationwide can publish technology requirements, and the Nanshan government subsidizes their fulfillment. Mechanism: 10 million RMB in demand β 5 million RMB in government support, of which 2.5 million RMB is an advance payment. This is an unprecedented level of pre-financing to stimulate practical robot applications.
"We see new letters added to model names every day. The 26 letters of the alphabet will soon run out. But for those of us coming from autonomous driving, there is no fundamental difference between these models β we've already traveled the path from divergence to convergence."
β Yu Kai, founder of Horizon Robotics (as recounted by Dr. Tan Weijia)
GBA's strategic bet: "Make hardware sufficiently safe and sufficiently cheap." If the hardware platform cost drops from $100K to $10K or even Β₯10K, the application market will explode. GBA occupies a unique position to achieve this goal thanks to the simultaneous presence of all three supply chain tiers.
"The EU AI Act is not a barrier to technology. It is a differentiated risk management system. If your AI product is classified as high-risk, you can still enter the market β provided you fulfill the obligations."
β Tang Yilin, AI R&D Manager, TΓV Rheinland
Tang Yilin presented a detailed breakdown of the EU AI Act, the world's first comprehensive artificial intelligence law. Key parameters:
| Risk Level | Description | Obligations | Examples for Robots |
|---|---|---|---|
| π΄ Prohibited | 8 prohibited practices (effective since February 2025) | Product modification or market entry ban | Social scoring, exploitation of vulnerabilities |
| π High Risk | Two categories: product safety + 8 areas | Risk assessment, CE marking, third-party audit | Robot safety, biometrics, emotion recognition |
| π‘ Specific Risk | Transparency requirements | User notification, machine-level watermarks | Voice assistants, robot chatbots |
| π’ Minimal | No mandatory requirements | None | AI office tools, spam filters |
Tang Yilin emphasized that the 6 roles have different obligations, and one manufacturer can occupy multiple roles simultaneously:
| Role | Obligations | Complexity |
|---|---|---|
| Provider (provider) | Full risk assessment, CE marking | π΄ Maximum |
| Product Manufacturer | Integration of AI into product = liability equals Provider | π΄ Maximum |
| Deployer (deployer) | Use in professional activities, without AI modification | π‘ Medium |
| Importer | Verification of provider compliance | π’ Low |
| Distributor | Ensuring documentation availability | π’ Low |
| Authorized Representative | Representation of non-European manufacturer | π’ Low |
A critically important warning for robot manufacturers: if you integrate someone else's AI into your product and enter the market as an integrated device, you are the sole responsible entity. Even if the AI component is provided by a third party, all compliance falls on you as the Product Manufacturer.
Section A: products for which the AI component automatically falls under high-risk AI Act requirements (including Machinery Regulation, toys, medical devices). Full third-party audit required.
Section B: sectoral legislation takes priority. If your product falls under Section B, AI requirements are minimal, satisfying sectoral norms is sufficient. This prevents double regulation.
As of May 2026, the transfer of MR (Machinery Regulation) from Section A to Section B is under discussion. If this happens, AI Act requirements for robots will be substantially simplified, it will be sufficient to meet the AI requirements of the MR itself. This is a potentially very positive development for the industry.
Tang Yilin identified the central problem for the humanoid robot industry: AI is deeply integrated into safety functions. Obstacle avoidance, path planning, speed and force control, many manufacturers report that "80β100% of these functions are implemented through AI." This means that the robot's AI system is not merely an additional feature, but a safety component subject to the strictest evaluation.
| Priority | Characteristic | Testing Approach |
|---|---|---|
| Tier 1 (Critical) | High PLr + deep AI integration + no alternative path | Full AI testing, possible software audit |
| Tier 2 (High) | High PLr + AI integration + possible alternative technical path | Priority testing, but lower than Tier 1 |
| Tier 3 (Auxiliary) | AI used for auxiliary functions, not for safety | Reduced testing priority |
The compliance process under MR includes: determining whether AI performs safety functions β if yes, high-risk AI Act requirements apply β Module B/C conformity assessment β declaration of conformity β CE marking β market entry.
Tang Yilin proposed a two-tier strategy: (1) Basic Compliance, a "horizontal" approach to obtain the entry ticket to the EU market; (2) Functional Verification, preparation for the inevitable tightening of software testing requirements. Companies that start preparing now will gain a competitive advantage when solid AI testing standards become mandatory.
"MR is not just a directive update. It is a fundamental shift: Safety includes Security. If your robot is hacked and causes physical harm β that is an MR problem, not just CRA."
β Fan Zong, TΓV Rheinland
Li Jie (TΓV Rheinland Certification Expert) presented a detailed breakdown of the new Machinery Regulation (MR), which will become mandatory on January 20, 2027 (adopted July 19, 2023, 42-month transition period, 42 months transition).
| Space | Definition | Robot Action |
|---|---|---|
| Maximum Space | Physical boundaries of the environment (room walls) | Free movement |
| Restricted Space | Accessible space accounting for furniture and obstacles | Normal navigation |
| Monitored Space | Zone monitored by sensors (LiDAR, cameras) | Object detection |
| Safety-Protected Space | Zone in which the robot starts decelerating upon human detection | Speed reduction |
| Protective Space | Zone closest to the robot β immediate stop upon human entry | Protective stop |
Li Jie noted a critical emergency stop problem: "When the emergency stop is pressed, the robot may fall and cause a new type of injury. This is unacceptable, a protective measure must not create new hazards. A solution exists, but requires deep engineering analysis of each scenario."
Fan Zong (TΓV Rheinland) presented a comprehensive map of cybersecurity requirements for robots in the EU:
| Regulation | Standard | Key Date | Essence |
|---|---|---|---|
| RED (Radio Equipment Directive) | EN 303 645 | Effective | Basic cybersecurity for radio devices |
| IDDA (IoT Delegated Act) | EN 18031 | August 1, 2025 | 14 dimensions: access control, authentication, data protection |
| CRA (Cyber Resilience Act) | EN 4xxxx series | Phase 1: September 11, 2026 Phase 2: December 11, 2027 | Phase 1: vulnerability management processes. Phase 2: full product compliance |
| MR (Machinery Regulation) | EN 50742 | January 20, 2027 | Security as part of Safety. Two routes: A (SRAL levels) or B (IEC 62443) |
Fan Zong emphasized the distinction between Safety and Security: "Safety is physical harm to people and property. Security is protection against unauthorized access. But in MR they are united: if a hack causes physical harm, that is already a Safety problem regulated by MR."
"Practically any robot with a programmable chip and digital interface (Bluetooth, WiFi, USB, Type-C) falls under CRA. We have checked β almost all of our clients' products pass this test. CRA is inevitable."
β Fan Zong, TΓV Rheinland
Strategic recommendation: start with IDDA (EN 18031), since its results easily map onto CRA and other standards. This minimizes costs when sequentially passing multiple certifications.
Liang Liubo (Byron), Senior Business Development Manager at TΓV Rheinland, presented an analysis of privacy threats specific to embodied robots:
The UK Product Security and Telecommunications Infrastructure Act (PSTI), which came into force on April 29, 2024, establishes three basic requirements for all consumer IoT devices:
| # | Requirement | Description |
|---|---|---|
| 1 | Unique Passwords | Prohibition of universal factory passwords. Each device β a unique password, changeable by the user |
| 2 | Support Period | Transparent indication of the period during which the manufacturer commits to issuing security updates |
| 3 | Vulnerability Reporting | Availability of a contact point and procedure for receiving vulnerability reports from researchers |
PSTI fines: up to Β£10 million or 4% of annual revenue, comparable to GDPR. The requirements seem simple, but non-compliance leads to catastrophic consequences.
Byron described multiple attack vectors: USB, WiFi, BLE, 4G/5G, GPS spoofing (coordinate substitution via signal generator), pseudo base stations for 4G. Key assessment methodology: OWASP IoT Top 10, the ten most critical IoT vulnerabilities.
"Robot cameras and microphones are constantly active β for navigation, for voice commands. Are facial data being collected? Voice data? Are they transmitted to a server? Without compliance, this is a time bomb."
β Liang Liubo (Byron), TΓV Rheinland
Byron gave a personal example: "I myself, knowing everything about cybersecurity, place my phone camera-up in the bathroom. Because the voice wake-word listens constantly, the unlocking camera scans the face constantly. Are these data being transmitted? That is a question of the manufacturer's conscience."
Types of personal data collected by robots: facial images, voice samples, geolocation, behavioral patterns, emotion biometrics, home environment data. Each type requires specific protection measures: from anonymization (face masking) to full encryption.
"Imagine: you're flying with a 30-kilogram robot from Oakland to Singapore. At security, they tell you β remove the battery. Removed? Great. Now check it in. But you don't have the UN 38.3 certificate. Result: you carry the robot on your back halfway around the world."
β Li Shilin, Vice President TΓV Rheinland North America (real case, May 2026)
Li Shilin presented a comprehensive map of requirements for entering the US and Canadian markets:
| Domain | Regulator / Standard | Requirements |
|---|---|---|
| Electrical Safety (Workplaces) | OSHA + NRTL | Mandatory certification through a recognized laboratory (TΓV is an NRTL) |
| Electrical Safety (Consumers) | CPSC | Certification through an accredited third party |
| Service Robots | UL 3300 | Safety of service and communication robots |
| Industrial Robots | UL 1740 | Robots and robotic equipment |
| Kitchen Robots | UL 3320 + 197 | Commercial; for domestic β UL 1026 |
| Electromagnetic Compatibility | FCC / IC (Canada) | With wireless module β FCC ID; without β SDoC |
| Laser Safety | FDA CDRH | Registration of laser products (LiDARs β usually Class 1) |
| Energy Efficiency | DOE / CEC (California) | Minimum standards for charging systems |
| Chemical Safety | TSCA + CP65 (California) | Heavy metals, toxic substances on surfaces |
| IoT Security | SB-327 (California) | Unique Passwords, encryption, firmware updates |
| Children's Privacy | COPPA | Children's data protection: parental consent, right to deletion |
| AI Compliance | FTC / SEC / ICC | Fragmented oversight: FTC β fraud, SEC β trading algorithms, ICC β media |
| Cyber Trust Mark | FCC (voluntary) | Standard NIST 8425, the Energy Star equivalent for cybersecurity |
Li Shilin noted that cybersecurity in the US is fragmented: at the federal level there are no mandatory requirements yet, but SB-327 (California) has de facto become the national standard. The FCC is preparing a voluntary Cyber Trust Mark (based on NIST 8425), it is expected that the absence of such marking will become a market barrier, even without a formal mandate.
Li Qin (Senior Certification Expert, TΓV Rheinland) presented a deep technical breakdown of robot functional safety:
"On June 1, in a Shenzhen park, a robot kicked a child during a demonstration. In the US, a robot charger caused a house fire β fortunately, the owner was not home."
β Li Qin, TΓV Rheinland (on real 2026 incidents)
PLr is the target functional safety level, determined by three risk parameters:
| Parameter | Designation | Value |
|---|---|---|
| S (Severity) | Severity of possible injury | S1 (minor) or S2 (severe, irreversible) |
| F (Frequency) | Frequency and duration of hazard exposure | F1 (rare) or F2 (frequent/prolonged) |
| P (Possibility) | Possibility of avoiding or limiting harm | P1 (possible) or P2 (virtually impossible) |
| Parameter | Threshold | Consequences |
|---|---|---|
| Collision Force | < 50N | Functional safety not required (no risk per standard) |
| Collision Force | > 50N | Functional safety required; level depends on force magnitude |
| Robot Weight (Tip-Over) | < 1 kg | No risk (level 1) |
| Robot Weight (Tip-Over) | > 1 kg | Risk level 2 or 3; Category B/C architecture mandatory |
| Factor | Designation | Description |
|---|---|---|
| Architecture Category | Cat. B/1/2/3/4 | B β basic, 1 β with reliable components, 2 β with diagnostics, 3 β with redundancy, 4 β with self-diagnostics and redundancy |
| Mean Time to Dangerous Failure | MTTFd | Not the component's service life, but the probability of a dangerous failure in the safety system |
| Diagnostic Coverage | DC | The system's ability to detect its own dangerous failures before they lead to an incident |
| Common Cause Failure Resistance | CCF | Protection against situations where a single event disables all redundant channels simultaneously |
Li Qin emphasized: PLr is the target (what is required), PL is the fact (what has been achieved). Verification includes hazard analysis (human, environment, equipment, energy), three protection classes (mechanical, electrical, fire), and strict emergency stop requirements (red marking, ergonomic accessibility, compliance with functional safety standards).
Key dates in chronological order:
Global context: Vietnam (April 2026) issued detailed AI rules, structurally mirroring the EU AI Act. Singapore, Korea, in progress. USA, fragmented approach: FTC (combating AI fraud), SEC (trading algorithms), ICC (content platforms). Legal precedent: lawsuit against Character.AI (2026) regarding formation of emotional dependency in a user.
"If robots start competing for real β other industries won't stand a chance. We have never seen a market larger than the robot market."
β Dr. Tan Weijia, Secretary-General, Shenzhen Robotics Association
The TΓV Rheinland GBA forum highlighted the fundamental transformation the robotics industry is undergoing. Key synthesis conclusions:
The industry is transitioning from "spectacle" (CCTV Galas, demo videos) to real production pilots. Chinese manufacturers have created a world-class robot "body." The next step, "hands" (dexterous manipulators with tactile feedback) and "brain" (reliable AI, certifiable to safety standards).
A thesis repeatedly echoed at the forum: compliance is not a barrier, but a passport. TΓV Rheinland certification (CE, NRTL, CB) reduces insurance risks, opens market access, and serves as a differentiator in the fight against dumping. Li Dong summarized: "Don't lie on the path to market, ensure compliance first."
The 2026 incidents, a robot kicking a child in a Shenzhen park (June 1), a charger fire in the US, a traveler forced to disassemble a robot at the airport, demonstrate that the absence of safety standards is the largest business risk, not an abstract regulatory problem.
| Market | Regulatory Status | Recommendation |
|---|---|---|
| EU | Maximum density: MR (Jan 2027), CRA (2026β2027), AI Act, GDPR, IDDA | Start with IDDA (EN 18031), reuse for CRA |
| UK | PSTI (2024), GDPR-like regulation | 3 basic requirements β low entry threshold |
| USA | Fragmented: OSHA/NRTL, CPSC, SB-327, FCC, FDA, COPPA | UL 3300 as core; prepare for Cyber Trust Mark |
| Canada | SCC, IC, provincial requirements | Certification through recognized bodies |
| Australia | C32 (PSTI equivalent) | Basic cybersecurity requirements |
| APAC | Vietnam (2026), Singapore, Korea β on the EU model | Monitoring; EU compliance as passport |
The GBA 2026 Forum captured a historic moment: Chinese robotics stands on the threshold of global expansion. Three pillars of success: engineering excellence (GBA supply chain), preemptive compliance (TΓV Rheinland as partner), and focus on real scenarios (the "Jiebang Guashuai" policy). Ahead, the 15th Five-Year Plan (June 2026), which will define the next five years of development. The window of opportunity is open, but narrowing with each regulation coming into force.
Although NuCypher technologies (Proxy Re-Encryption, PRE) and other advanced cryptographic methods such as ABE (Attribute-Based Encryption) and FHE (Fully Homomorphic Encryption) were not directly discussed at the forum, they represent the next frontier in embodied robot data protection. PRE enables delegating access to encrypted data without revealing keys, critically important for robot data processing pipelines. ABE provides granular attribute-based access control. FHE potentially enables processing robot data without decrypting it, solving the fundamental privacy problem raised by Liang Liubo (Byron) in the context of constantly active cameras and microphones.