The "Disruptor Sample" at WRC 2025: The Born-Mature Luming Robot

At the 2025 WRC site, people swarmed around the exhibition booths, observing humanoid robots of various shapes. There were familiar faces like Unitree, Zhipu, and Galaxy. In contrast, the robots of one company didn't initially attract much attention.

When approaching their booth, the robot was dancing to the music. The dance was flexible and coherent without any glitches, and even the rhythm of the steps matched the music. Then, the staff at the booth made the robot lie down, and the booth started timing. The modules of the robot's hip and knee joints instantly exerted force, and it bounced up from the ground in just one second, standing steadily without any extra movements.

The lightning - fast standing - up movement made this robot company, making its debut at the WRC, instantly catch people's eyes.

This company is Lumos Robotics. It was founded last September, but in less than a year, it has almost completed the preparations that other humanoid manufacturers took several years to achieve during the same period.

Considering the industry situation, currently, domestic humanoid robot manufacturers are competing on multiple paths. There are companies like Ubtech, Unitree, and Zhipu that have grown natively from the robot track; Topsstar and EVATEC have made cross - border transformations from the industrial field; technology giants like Huawei are deploying on the AI side; and Xiaomi, XPeng, and Tesla are extending their humanoid businesses with the resources of automobile manufacturers.

Although the competition seems fierce, the overall industry is still in a critical development period. The new forces in the robot industry have not yet formed a stable order - fulfillment loop, and the cross - border transformation manufacturers are limited by the weak accumulation of embodied intelligence technology.

And the team genes of an enterprise often make it start from a completely different starting point.

Yu Chao, the founder of Lumos Robotics, graduated from Tsinghua University, obtaining degrees in mathematics, energy and power, and aerospace. He has 10 years of experience in the R & D and industrialization of embodied robots. During his tenure as the head of the Embodied Robot (Magic Lab) at Dreame, he completed the team building and the R & D and mass - production projects of Xiaomi Cyberdog and Dreame's quadruped robot. He led the team to finalize the embodied software and hardware framework and pioneered the world's first electric - driven embodied robot capable of a backflip. Other core members of the company include the former core team of Dreame's embodied robots and other senior practitioners in the robot industry. The R & D team includes many doctors from top universities.

Currently, the team has launched products such as the full - size humanoid robot Lus, the robot Hercules MOS for industrial logistics scenarios, the small - size humanoid NIX for entertainment and household fields, as well as multiple integrated joint modules and visual - tactile sensors. It completed three rounds of financing in the first half of this year, directly shortening the long cycle of "pilot - verification" and entering the "pre - mass - production lock - in" stage in advance, successfully squeezing into the final sprint.

This high starting point and rapid progress are precisely due to the Lumos team's 10 - year industrial experience and mass - production accumulation of over a thousand units.

Get up in One Second, Breaking the Technical Limit of Humanoid Robot Motion Control

Currently, the "size differentiation" in the humanoid robot market is obvious. Mainstream products like Unitree G1 have a height concentrated in the range of 120 - 130 cm. Although they can perform basic actions such as dynamic standing up and stick - dancing, due to their size, they have natural shortcomings in adaptability to the human environment. The tools and operation scenarios in daily human use are all designed according to the adult proportion. Medium and small - size robots are often limited by the inability to "reach high places" or "insufficient operation range" when grasping high - placed items or adapting to complex industrial operating tables, thus reducing the applicable scenarios.

In the past, the reduction in size was more of a compromise to technology. There is an intuitive description in the humanoid field: "For every 10 - centimeter increase in height, the R & D difficulty doubles." An increase in height directly leads to a higher center of gravity. The center of gravity of a 1.6 - meter robot is 33% higher than that of a 1.2 - meter robot, and the stability threshold during dynamic balance is significantly reduced. "Growing taller" also affects joint drive, causing a doubling of the robot's torque requirements and impact loads. At the same time, the extended limbs increase the robot's moment of inertia, requiring a reconstruction of the motion control algorithm to handle the strong inertia and complex dynamic balance problems. The system energy consumption caused by the large body cannot be ignored, which affects the robot's working hours.

These challenges make full - size robots a "tough nut to crack".

In scenarios that test extreme performance, this technical gap is further magnified. In the world's first humanoid robot fighting competition in May this year, the rule was that "if a robot fails to get up within 8 seconds after falling, it is judged as KO". These 8 seconds also reflect the limitations of humanoid robots in emergency response speed and dynamic balance control.

Against this background, Lumos Robotics launched the world's first full - size humanoid robot LUS2 capable of "ejecting up in 1 second" this year.

LUS2 is designed in an adult proportion of 160 cm. It can complete long - range actions without remote control, achieve a 2 - hour high - intensity sports endurance, and can be applied in multiple scenarios such as science education, guided tours, and entertainment.

The above - mentioned performance is based on the advantages of Lumos Robotics' overall motion control architecture. According to the company, LUS2 is equipped with an ultra - high - speed dynamic balance control system. Through the bionic posture algorithm trained by reinforcement learning, it can complete the decision - making of center - of - gravity migration in 1 millisecond, equivalent to a dynamic adjustment ability 30 times faster than the human reflex nerve. It is also equipped with an IMU + joint torque + joint angle perception system, which allows the robot to generate a recovery trajectory in any posture. To improve the real - time response speed, LUS2's real - time control architecture can achieve microsecond - level synchronization of sensor data collection, motion planning, and motor drive. Moreover, when a single joint is suddenly overloaded, it can still complete the action through the torque redistribution of the remaining joints.

In addition, in the bio - like joint drive system, Lumos has self - developed an actuator with a peak torque of 380 Nm and a torque density of 233 Nm/kg. With an instantaneous response ability of 0.2 seconds, it is equivalent to 1.8 times the lower - limb explosive force of an adult male. The exoskeleton - type architecture maximally protects the internal transmission mechanism and absorbs the ground impact force through the passive compliance mechanism.

The deeper advantage of this technical architecture lies in its technical reusability. Each module of the above - mentioned system can be directly migrated to actions such as running and handling, truly enabling humanoid robots to step out of the "laboratory".

Three Products in One Year, a Crazy Iteration in R & D

This year, humanoid robots have appeared in the public eye as a high - frequency term. There are endless competitions and demonstration booths. Even at this WRC, 50 humanoid robot manufacturers competed on the same stage. The rapid market update is not only reflected in the frequent breakthroughs in technical parameters but also in the rapid fission of various scenario requirements. Behind this high - speed iteration is a severe test for enterprises' product strategies and R & D systems in the industry.

Among them, focusing on a single product or technology is one direction for the development of humanoid manufacturers. Undoubtedly, focusing on a single product has certain advantages for specific scenario applications, which can help enterprises quickly implement and achieve mass production in a single scenario. However, the disadvantages of this path are also obvious. In the early stage of humanoid robots, PMF (Product Market Fit) needs to be co - created with users. A single robot can hardly meet the requirements of multiple scenarios. Therefore, more humanoid companies tend to choose the strategy of "laying eggs along the way", verifying multiple technical paths through diversified products simultaneously to avoid missing the opportunity of technical iteration by betting on a single solution. To support this strategy, a company needs to have the ability to quickly develop and iterate multiple products.

Lumos is one of the few domestic companies with full - stack technical capabilities including the robot body and brain. Through the innovation of software - hardware integration, it has fully self - developed high - cost and high - difficulty core hardware such as high - performance integrated joints and visual - tactile sensors. In terms of the brain, the company has built a technical foundation of an operation large model and a whole - body motion control model. Based on its "full - stack R & D capabilities", Lumos has established a modular R & D system similar to Lego building blocks. The modular reuse enables Lumos to achieve parallel development of multiple product lines while maintaining technical consistency.

This is Lumos' competitive strategy. They believe that in the chaos of technical routes, only a technical system capable of supporting the rapid iteration of multiple products can grow into an industry infrastructure.

In less than a year, Lumos has launched three complete robots and multiple component products. In the early stage of the industry, this iteration speed is undoubtedly astonishing.

In terms of products, Lumos' product categories are diverse, and the application scenarios are also very extensive. In addition to the full - size humanoid robot Lus series mentioned above, the robot Hercules MOS for industrial logistics scenarios has a dual - arm load capacity of 50 kg, which is particularly suitable for dual - arm operations in high - load industrial scenarios. According to industry - disclosed data, the maximum dual - arm load capacity of most dual - arm wheeled robots on the market is within 20 kg, forming a gap in capabilities. The MOS series robots developed by Lumos precisely fill this gap with a maximum load capacity of 50 kg, meeting the real needs of the scenario. In addition, the 80 - cm - tall small humanoid robot NIX is targeted at the family companionship and commercial service fields, being lightweight, highly stable, and more cost - effective. Other embodied intelligent hardware series self - developed by the company, such as high - performance integrated joint visual - tactile sensors, grippers equipped with visual - tactile sensors, 7 - degree - of - freedom data acquisition arms, and data acquisition solutions, are also on sale.

Co - building Scenarios with Leading Customers in Various Industries, Cracking the Core Code of Scale - up

Humanoid robots can't operate without real - world data. To truly achieve large - scale implementation, in addition to solving problems such as the maturity of hardware and components and large - scale cost reduction, the biggest limiting factor is high - quality real - machine data, which must be generated through interactions in real scenarios. Therefore, co - creating with customers in various industries has become the unanimous choice of humanoid robot manufacturers.

Leading domestic and foreign enterprises have deeply bound themselves to real - world scenarios and carried out case studies. For example, Figure AI's Figure 02 entered the BMW factory; Tesla's Optimus "started working" in its battery factory; and Ubtech's Walker S entered an automobile factory to work.

The quality of Lumos Robotics' scenario partners is also high - level. In the past six months, Lumos Robotics has successively reached strategic cooperation agreements with companies such as Dematic Technology (SH688360), a smart logistics solution provider, COSCO Shipping Group, a global shipping giant, and Mitsubishi Electric, a global leading digital factory solution provider. They will jointly develop solutions for embodied intelligence in scenarios such as flexible quality inspection, logistics, and port shipping and then promote them in the global market.

The core reason why Lumos can win over many leading customers in a short time is its full - stack technical capabilities with both software and hardware and its highly distinctive product capabilities, which can meet the real needs of scenarios. Take the common handling scenario in the industrial and logistics fields as an example. Industry data shows that 90% of material transfer requirements are carried by 600×400×250 mm plastic turnover boxes. Especially in production - line logistics and warehousing logistics, the weight of the materials in the boxes is often over 30 kg. The load capacity of most dual - arm wheeled robots on the market is only within 20 kg, creating a gap in capabilities. The MOS series robots developed by Lumos precisely fill this gap with a maximum load capacity of 50 kg, meeting the rigid needs of the scenario.

As robots continue to "work" in scenarios such as industry, logistics, education, and entertainment, each set of accumulated data makes the robot's brain better understand the needs of logistics scenarios. This evolution based on real data is the key for Lumos Robotics to achieve large - scale implementation in various scenarios.

Conclusion: On the Eve of the Explosion, Lumos is Ready

According to GGII's prediction, the global sales volume of humanoid robots is expected to reach 12,400 units in 2025, with a market scale of 6.339 billion yuan; by 2030, the sales volume will be close to 340,000 units, and the market scale will exceed 64 billion yuan; by 2035, the market scale will exceed 400 billion yuan.

Standing on the eve of the explosive growth in 2025, we still have many questions and uncertainties about humanoid robots: Will there be a disruptive iteration in the technical route? Which scenarios can truly support the commercialization loop? When will the critical point of hardware cost reduction arrive? In these unknowns, the "maturity" of an enterprise's starting point is particularly crucial. Those players with deep technical accumulation, mature product thinking, and commercialization capabilities from the very beginning are more likely to quickly find a definite path in the chaos.

As an investor said, Lumos Robotics is the fastest - growing embodied robot company in 2025. Whether in terms of product performance, iteration speed, or commercialization progress, it far exceeds other start - up companies in the same period. Lumos Robotics' "born - mature" performance fully verifies the team's deep historical accumulation and has the opportunity to become the most anticipated embodied intelligence start - up company after Unitree and Zhipu.