The journey of humanoid robots from mere concepts to tangible realities is accelerating at an unprecedented pace.

On February 11, 2025, Yushutech, a technological firm based in Hangzhou, launched two new humanoid robot models on its official JD.com flagship store, labeled as Unitree H1 and G1. Just a day later, the G1 model, priced starting at 99,000 RMB, sold out upon its initial release.

In automated factories, these humanoid robots have begun to showcase tangible training resultsAccording to representatives from Ubtech Robotics, significant progress had been made with their Walker S1 model during training sessions at BYD Automotive's factory, where efficiency doubled and stability improved by 30%. Optimization efforts are ongoing, with plans for large-scale deliveries projected by the second quarter of 2025.

The same representative mentioned that at Geely, the Walker S1 completed the second phase of testing at the Zeekr factory in NingboBased on the successful initial phases, Geely has assigned Walker S1 to conduct further training focused on testing charging gun insertions and material handlingAt Foxconn, the Walker S1 has completed its first phase involving logistics scenarios, and further tests will expand into more areas during the second phase.

The humanoid robotics industry has evolved significantly since Boston Dynamics unveiled the Atlas humanoid robot in 2013. From the flashy era of Atlas performing backflips to the practical applications witnessed now, like the Walker S1’s precision in managing automobile wiring harnesses, the industry stands at a pivotal juncture, ready for commercialization.

The last significant challenge before widespread commercialization of humanoid robots lies in achieving mass production.

Mass Production

On February 6, 2025, Tesla updated multiple job postings related to humanoid robots on its official website, seeking engineers, process supervisors, and production managers, clearly marking them as positions for the “Tesla Bot.”

These positions are based in Tesla's Fremont factory in California, one of its largest manufacturing bases

Advertisements

In 2024, Tesla plans to deploy several of its self-designed Optimus robots there, primarily responsible for tasks such as material handling, battery sorting, vehicle body welding, and parts installationThe company emphasizes its objective of mass-producing humanoid bipedal robots that can autonomously handle repetitive and tedious tasks in manufacturing and logistics.

Previously, CEO Elon Musk unveiled ambitious plans for the production of the Optimus humanoid robot, aiming to manufacture 10,000 units in 2025 and commences delivery to companies outside Tesla by the latter half of 2026. Musk estimated that Optimus harbors the potential to generate over $10 trillion in revenue.

During the quarterly earnings call on January 29, 2025, Musk expressed an exceptional urgency and optimism regarding the production ramp-up of Optimus: “This is an exponential growth curve, shifting from having nobody using humanoid robots to an inundation of these robotsWe will forever exist in a state of ‘not doing enough.’ Even if the robots are expensive, demand will never be an issue.”

He highlighted that Tesla's current production line is achieving approximately 1,000 units per month for Optimus, with ambitions for subsequent lines targeting outputs of 10,000 and eventually 100,000 units monthly.

In China, several humanoid robotics companies have also announced their mass production plans.

On January 17, 2025, Leju (Shenzhen) Robotics Technology Co., Ltd. celebrated the delivery of its 100th full-sized humanoid robot at a ceremony held at the Beijing Automotive Group's off-road vehicle division, marking a new milestone in its production journey.

Earlier, on January 6, 2025, ZhiYuan Robotics rolled out its 1,000th general-purpose humanoid robot, with a reported total production of 731 biped humanoid robots.

“Ubtech's industrial humanoid robot Walker S has already received over 500 intention orders from automotive manufacturers and is currently in a critical stage of commercialization, with plans for large-scale delivery anticipated by the second quarter of this year,” disclosed an official from Ubtech on January 10, 2025. Previously, Ubtech had conducted humanoid robot training across various notable automotive factories, collaborating with firms like Dongfeng Liuqi, Geely, FAW-Volkswagen, Audi FAW, BYD, BAIC New Energy, and Foxconn among others.

The SE01 robot from Shenzhen Zhongqing Robot Technology Co., Ltd. garnered significant attention early in January 2025 as videos surfaced of it walking in urban settings, showcasing a walk as natural as that of a human.

On January 22, during a visit to Zhongqing Robotics, co-founder and marketing head Yao Qiyuan disclosed plans for multiple robot models to achieve mass delivery by 2025, stating, “In the humanoid robot sector, we have transitioned from ‘0 to 1’; now some companies are advancing from ‘1 to 100’, even moving toward ‘100 to 1,000.’”

This transition from technical validation to a commercially viable cycle marks a significant leap for the humanoid robot industry

Advertisements

Mass production not only demonstrates the market’s inherent potential but is also critical for the whole sector, as it facilitates cost reductions through economies of scale while optimizing design models and structures.

However, Yao Qiyuan pointed out that producing 100 to 1,000 units is still a form of low-scale production and does not yet push the industry toward a transformational iteration or structural adjustment.

A marketing director from a robot company in Shenzhen shared insights about the humanoid robot production timeline, delineating the evolutionary process into two key stages.

The first stage emerges when production hits 100,000 units, focusing primarily on verifying technological feasibility over cost controlThis stage targets substituting about 10% of long-tail scenarios within the manufacturing sector—non-standardized, low-repetitive tasks—where success relies heavily on the hardware's motion control stability and continuous production assurance, as well as software optimization for executing specific tasks under controlled environments.

The second stage signifies the commencement of massive production once outputs surpass 1 million unitsHere, hardware costs must drop below the average annual labor cost for a worker, and software systems must reach breakthroughs in cross-scenario generalization capabilities, enabling the application of robots beyond intelligent manufacturing into service sectors and home environments.

Musk indicated in the earnings call that by time production hits 1 million units, the price of the Optimus robot could potentially reduce to $20,000.

Nevertheless, the current market scale for humanoid robots remains unclear, and mass production will take time

Advertisements

Dongwu Securities, in a recent report, projected that due to high demand from downstream clients for testing and research, domestic humanoid robot sales in 2025 could approach 10,000 units.

Research from the High-tech Industry Research Institute (GGII) indicates that the global humanoid robot market size will be approximately $1.017 billion in 2024, potentially reaching $15.1 billion by 2030, with annual sales projected to grow from 11,900 units in 2024 to 605,700 units by 2030.

Hardware and Software Challenges

Industry insiders reveal two crucial challenges—both hardware and software—that humanoid robotics must overcome to scale up production successfully.

The hardware challenges predominantly revolve around three core components: drive units (motors + gearboxes + lead screws), sensing systems (vision/touch/force), and control modules (chips), with a notably bottleneck in manufacturing planetary roller lead screws, which are pivotal for precise motion conversion.

As key transmission components for linear joints, each humanoid robot requires between 10 and 14 planetary roller lead screws, which account for 20% of the joint module’s value and about 5% to 8% of the robot’s overall valueManufacturing these screws involves high precision processes that rely on specialized machinery and alloys, historically dominated by companies such as GSA from Switzerland and Rollvis from Germany.

“To achieve a production level of 1 million humanoid robots, the demand for these lead screws could reach tens of millions, and current production levels simply don’t sufficeBefore humanoid robots entered the market, lead screws were typically used in niche applications, with a market size of only a few billion yuan,” a researcher from a mid-sized automation company in Shenzhen explained.

According to data from Mitu Consulting, the Chinese lead screw market (excluding humanoid robots) was only about 2.57 billion yuan in 2023.

Notably, several domestic manufacturers are beginning to establish production lines for planetary roller lead screws.

On January 3, 2025, Hangzhou Xinjian Mechanical and Electrical Drive Co., Ltd. held a groundbreaking ceremony for its project to produce up to 1 million planetary roller lead screws for humanoid robots annually, with an investment of 2.6 billion yuan aimed at establishing a smart manufacturing line in two phases, with the first phase involving 1 billion yuan.

On January 15, 2025, Ningbo Zhenyu Technology Co., Ltd. announced it had built a semi-automated production line for planetary roller lead screws, with a daily capacity now reaching 50 sets

The company has also established an integrated screw testing laboratory to verify critical parameters such as screw lead accuracy and transmission efficiency.  To meet market demands, a second semi-automated production line is under construction, expected to be operational in the first quarter of 2025. The company plans to develop a fully automated line for processing, testing, and assembly to enhance production consistency.

“Our company believes that the future of the robotics supply chain could rival that of the new energy vehicle industryAt the outset, many parts manufacturers were focused on proving their capabilities; as the industry matures, we must verify our ability to maintain quality while reducing production costs,” a management representative from Zhenyu Technology remarked about the budding opportunities with the advent of humanoid robot mass production.

“Moving to the next step for humanoid robotics requires not only continued refinement of the supply chain but also ongoing advancements in algorithms to enhance functionalities and broaden application scenarios,” added Yao Qiyuan.

According to Yao, humanoid robots heavily rely on three core components: the brain, the cerebellum, and the body itselfThe brain is responsible for high-level decision-making and intelligent processing, while the cerebellum focuses on motion control, and the body serves as the physical structure to execute specific tasks.

The surge in popularity of the DeepSeek AI model since 2025 has highlighted the potential applications of large AI models within the cognitive framework of humanoid robotsThese models enhance generalization capabilities through pre-training and tuning, especially as parameters reach the billions, enabling robots to demonstrate complex thought processes and adapt to varied task environments. “However, the cerebellum and body remain significant bottlenecks in current developments,” Yao noted.

Yao further explained that the technological demands for the humanoid robot cerebellum are centered around its motion capabilities and robustness in navigating complex terrains—qualities essential for consistent operation amid various uncertainties and disruptions

Current motion control algorithms and systems face considerable challenges, particularly in achieving coordinated, nuanced movements across the entire robot bodyAdditionally, crucial issues persist concerning the speed-load characteristics, low drift, noise reduction, energy efficiency, and self-healing capabilities of the cerebellum’s technology.

To enhance operational performance, a high-fidelity modeling and simulation approach is needed for the cerebellum system, alongside multi-body dynamics modeling and real-time behavior control, which are vital for achieving biomimetic locomotion and self-learning in coordinated movements.

The research phase for humanoid robot cerebellum systems is currently more challenging than for the brain, primarily due to the scarcity of training data. “Real-world data is crucial for humanoid robot manufacturersIf each robot entity must gather its own real-time data, with so many different scenarios, it’s impractical for even the most powerful company to ‘consume’ all that data,” Yao stated.

Insiders have highlighted that controlling robot movement necessitates navigating dynamic and intricate physical environments, differing from the handling of data like images or textRobots must complete movement tasks in the real world, needing the cerebellum to account for various environmental variables, such as terrain, obstacles, object interactions, and diverse operational requirementsEach movement and decision carries potential varied feedback, making traditional simulation data inadequate to encompass such high-dimensional, diverse contexts.

During the interviews and research processes, it became evident that the training of the robot’s cerebellum is based not purely on static data analysis, but significantly relies on the robot's interactions with the environment during actual operations

Capturing relevant data necessitates interaction across varied environments, monitoring even the tiniest movements of the robot and the feedback received from its surroundings.

Compared to image recognition or voice parsing, action data for robots is notably complex and heavily reliant on specific hardware configurations, movement strategies, and control system adjustments, resulting in a relative scarcity of training data for motion control systems.

The Battle for Form Continues

During early 2025, when visiting various robotics companies in Shenzhen, it became apparent that not all firms share a bullish view on the potential of humanoid robots.

A marketing head from a robotic arm company expressed that humanoid robots might be better suited for domestic settings, while in industrial contexts, there’s little rationale for adhering strictly to humanoid formsSimilarly, a market head from a company focused primarily on vacuum cleaning robots stated their belief that the humanoid sector today lacks commercialization readiness, and thus they have no intentions of entering the market.

The reasoning from the robotic arm company's market head provided insights into robotic forms: “Quadrupedal or wheeled robots can travel at speeds two to three times quicker than humanoid robots over flat terrain, cutting energy consumption by more than 50%, while also reducing the complexity of control algorithms significantlyNon-humanoid robot hardware costs typically represent one-third to one-fifth of humanoid robotsIn logistics operations, about 70% of tasks can be addressed effectively with fixed robotic arms or automated guided vehicles (AGVs), minimizing the practical advantages of humanoid robots in highly structured environments.”

The ongoing debate regarding robotic form essentially encapsulates the deeper contest between technological pathways and commercial viability.

Yao Qiyuan remarked that the core value of humanoid robots lies in their general usability and environmental compatibility

Advertisements

Advertisements