The upcoming CES 2025 has already set the stage for a remarkable showcase of robotics and artificial intelligence (AI) technologiesThe excitement surrounding domestically-produced robot prototypes has reached a fever pitch, particularly following NVIDIA's founder and CEO Jensen Huang's assertion that we are on the verge of experiencing a transformative moment akin to the "ChatGPT revolution" in the realm of general robotics.

This surge in interest can be attributed to two primary factorsFirstly, the rapid evolution of core technologies, such as advanced AI models, is accelerating the practical application of robotics in various sectorsSecondly, the burgeoning demand for workforce automation has prompted multiple entities to get involved in deploying robots to handle an array of tasks.

Many local manufacturers have accumulated extensive international experience, especially in the field of robotic vacuum cleaners, which continue to evolve and expand their capabilities

Advertisements

Furthermore, more sophisticated types, such as humanoid robots and robotic dogs, are steadily entering the public consciousnessThis indicates a pivotal turning point, with 2025 being heralded as the year when humanoid robots will enter mass production.

During CES, Tesla's CEO Elon Musk elaborated on his plans to scale up production of the company's humanoid robot, named Optimus, stating that they aim to begin mass production in 2025, increasing output tenfold by 2026, with a target of producing between 50,000 to 100,000 units by 2027. This signals the start of a new race fueled by large AI models that promise to redefine technology applications.

Could we be on the cusp of the "ChatGPT moment" in robotics? At CES, journalists observed that the exhibits presented by domestic manufacturers spanned numerous applications, including both household and industrial use casesEven robotic vacuum cleaners, which have already established themselves in international markets, are experiencing continuous iterations leading to enhanced value propositions.

For instance, Doumi Technology has built upon traditional robotic vacuums, incorporating bionic multi-joint robotic arms that supposedly allow for grasping objects weighing between 400g and 500g within a 30cm spatial range

Advertisements

Once grasped, these objects can be returned to their designated places through the input from AI modelsThe addition of bionic legs also enables these robots to navigate obstacles over 4cm in height.

Furthermore, robotic vacuums have potential to evolve into lawnmowers, catering specifically to overseas market demandsStaff at the Ninebot booth explained that by utilizing visual and RTK technologies, they could ensure reliable positioning; their exhibited lawnmower can cover a larger operational area with significantly enhanced efficiency.

Companion robots are another hot topic of interestThe Chinese company Mengyou Robotics introduced Ropet, branded as the next-generation AI robotic petAs of the time of this writing, the project has received support from 1,012 backers on Kickstarter, raising a staggering HKD 2,142,800, far surpassing its initial crowdfunding goal of HKD 10,000.

A robot from Yushu Technology, which was entirely pre-ordered within half a day, marks a notable achievement as it is one of the first domestic firms to bring the price of large-scale robots below 100,000 yuan

Advertisements

Their exhibit showcased products like robotic dogs that can be used in familial settings and diversified applications including industry and emergency scenarios.

Humanoid robots drew immense crowds during the show as they engaged in frequent interactive sessions with attendeesAt Qualcomm's booth, "Ultra Magnus," a multimodal AI humanoid robot, designed around Qualcomm's SoC platform, was frequently conversing with visitors while delivering beveragesSon Xiaogang, CEO of AGASAI, expressed aspirations to leverage "Ultra Magnus" across industrial manufacturing, commercial services, and companionship sectors.

Musk's updates regarding the production timeline of Optimus drove area-wide attentionFollowing their plan, mass production is set to kick off in 2025, with a tenfold boost in production expected by 2026, aiming to create between 50,000 and 100,000 humanoid robots by 2027. This signifies Musk's belief in humanoid robots' capacity to enter critical segments of commercial operations.

Reflecting on the current wave of robotic innovation, the recent advancements capitalize on years of iterative evolution

• Definition and Types of International Capital Flows
• Slowdown in Indian Manufacturing Hinders Growth
• BYD Shares Rise While Tesla Stock Dips
• Gold Surpasses $2,700 Threshold
• Latin America Faces Urgent Call for Structural Reforms

Particularly pivotal has been the integration of AI models, which have significantly broadened the applicability of robots, resulting in the appearance of increasingly intelligent robotic productsTherefore, Huang's mention of an imminent "ChatGPT moment" in the robotic domain seems justifiable.

Yu Yiran, Executive Director at CIC Consulting, expounded on this development, citing technologies like deep learning, generative adversarial networks, and multimodal AIs as crucial components pushing the boundaries of AIThis enables robots to gain stronger autonomous learning capabilities alongside improved adaptability and comprehension of human commandsConsequently, the transition from mechanical devices to intelligent agents is becoming plausibleThe release of models such as Google’s RT-2 exemplifies this potential, as they can output actions without invoking other models, showcasing the power of end-to-end models.

Moreover, according to IDC's research manager Li Junlan, the shift in population structure pointing towards a "humans-replaced-by-machines" market demand is evident

The maturation of the robotics supply chain, plummeting costs, guiding national policies, and capital investments have also emerged as vital contributors to the current wave of interest surrounding robotics.

Nevertheless, challenges loom on the horizonMany industry professionals have indicated that as more data is accumulated from the physical world, the availability of real data for training models might be nearing depletionThis could delay the iterative advancement of large models and impede the continuous diversification of application scenarios.

In a response to the existing developmental gap, NVIDIA unveiled a physical world model at CESDubbed the Cosmos World Foundation Model, this empowers developers to generate vast amounts of physics-based synthetic data for training and evaluating their existing modelsIt enables developers to refine this framework into tailored models to best suit their workflow.

Huang stated, “Like large language models, the world foundation model is fundamental to the development of robots and autonomous vehicles

We created Cosmos to democratize physical AI, giving every developer access to universal robotics technologies.” The early batch of users for the Cosmos framework includes 14 robotics and automotive companies, with domestic firm Xpeng planning to utilize it for robotic innovations.

Yu explained that the Cosmos world model offers an integrated AI framework for the robotics sector, drastically minimizing R&D costs and time, thereby accelerating the commercial rollout of robotic technologies"Cosmos can create highly realistic virtual worlds where robots can engage in reinforcement learning, consistently experimenting with various actions and strategies while adapting their behaviors based on feedback for optimal decision-makingThis model-based approach expedites the speed at which robots learn while reducing the costs and risks associated with trial and error in real-world scenarios

Additionally, it facilitates cooperative tasks among robots from different fields, enhancing task execution efficiency while breaking new ground across industrial applications.”

Moreover, the rise of traditional companies that historically focused on sectors like automotive and telecommunications are now displaying increasing interest in roboticsSome manufacturers recognize the practical value of robotics for industrial applications, while the existing capabilities from their supply chains can be adapted to this new realm.

Notably, in 2024, Boston Dynamics, a veteran in the robotics space, is shifting away from hydraulic systems in favor of electric solutions, aligning with trends toward electrifying vehicle enginesLi Junlan noted that automotive and smartphone manufacturers possess inherent supply chain advantages within the robotics market“Building on smart driving technology, automotive companies already boast a supply chain foundation for precision components, such as sensors, control systems, and actuators, owing to their AI-related stature in real-time control and automation,” she said

“Mobile manufacturers also excel in miniaturization, high-performance computing, and precision manufacturing.”

However, Yu highlighted that while there is an evident first-mover advantage within electric vehicle manufacturers for developing purely electric robots, several technological barriers still exist"From a mechanical standpoint, the power and transmission systems of electric vehicles and robots share similarities, making it likely that motor control techniques will see quicker adoption in roboticsThe advanced visual techniques, high-definition radar, and multimodal sensors applied in electric vehicles offer reusable technology foundations for humanoid robots’ environment perception and interaction," she mentionedYet, the intricacy of AI technologies remains challenging to transfer.

“Although electric autonomous driving pertains to AI, its focus primarily lies in route planning and obstacle recognition in relatively contained scenarios, obtaining data mainly from road conditions and traffic laws

In contrast, robots must maneuver through more unpredictable human environments, incorporating various autonomous perception, planning, and execution tasks that require substantial data retraining,” she stated, indicating that electric vehicle manufacturers need to dedicate additional resources to R&D and innovation.

However, a noticeable trend suggests that the robotics sector is focused on specific applications, limiting the scope for widespread commercial successEven with Optimus demonstrating adept capabilities within factory environments, it is evident that it serves particular functions with narrow objectivesA broader mainstream acceptance of robots may take more time to cultivate.

Recently, Hu Baishan, an executive vice president at Vivo, provided insights, suggesting that achieving the dexterity and decision-making prowess envisioned for robots may require anywhere from ten to fifteen years

He emphasizes a phased implementation approach, proposing incremental product cycles for development iterations.

Consequently, while 2025 will mark a symbolic year for humanoid robot mass production, the expanse of developmental advancements must still progress before we realize widespread commercialization.

Chen Jun, Deputy General Manager and Chief Analyst at Sigmaintell, shared that the global shipment of robots is projected to exceed 47 million units in 2024, with an expected climb to nearly 60 million units by 2025, maintaining an annual growth rate of 18%.

“Service robots command the lion's share of the market, constituting around 80%, with approximately 70% of them dedicated to cleaning applications, such as robotic vacuums and window-cleaning robots,” he elaboratedObservations indicate that the total quantity of humanoid robots remains minimal, with projections estimating fewer than 4,000 units in 2024 and around 5,000 for 2025, indicating they remain in an initial growth phase.

Yu pointed out that the primary challenges hindering market deployment of robotics arise from high costs and the need for stability and reliability in technology.

“The Chinese market has a lower price tolerance for robots compared to international markets, causing limited profit margins for domestic firms,” she added, noting that robot prices are still considerably high

Tesla's Optimus Gen2 is currently estimated at $50,000 to $60,000, with post-mass production costs anticipated to drop to $20,000 to $30,000. Due to pricing constraints, there is a propensity for companies to seek foreign markets for better negotiating powerMoreover, several technical bottlenecks still constrain commercialization, including data scarcity, model reasoning efficiency, and reliability that need improvement.

The industry currently categorizes the progression of humanoid robots into three dimensions: the brain, the small brain, and the limbsIt becomes evident that large models significantly contribute to the development of the robot's brain and small brain while acknowledging that there’s much room for iterative improvement.

Li noted that at present, the “brain” remains the least developed area among humanoid robotsIDC research suggests that users express a pronounced desire for improvements in the robots' autonomous perceptual capabilities — the core of cognition and decision-making.

“The advantageous impact of large models on enhancing the robot’s brain is clear, with multimodal large models emerging as a critical direction

Future expansions will need to incorporate multimodal perception at the physical level (across vision, hearing, touch, inertia, and localization). Simultaneously, self-learning and iterative enhancement abilities also require further exploration,” she emphasizedIn terms of the small brain, additional optimizations for dynamic balance control, complex movement strategizing, and collaborative autonomous learning are crucialThe limbs demand breakthroughs in tactile feedback, soft material and structural design, and efficient energy management.

Furthermore, Yu elaborated that in prior developmental phases, humanoid robots have generally focused on the hardware and control elements, concentrating on establishing proficiency in their execution modulesConsequently, growth in the brain and small brain dimensions correspond to fast-evolving artificial intelligence realms, notably large-sensor models, which pose developmental challenges due to the lack of datasets