Alibaba

ByteDance

CATL

Huawei

Meituan

Tencent

Xiaomi

Most Followed

KrASIA Connection

Market Watch

Earnings Reports

Venture Roundup

China's Auto Sector

100 AI Creators

View All

Collections

News

Features

Insights

Ecosystem

Press Releases

DeepSeek

Li Auto

Luckin Coffee

Mixue Bingcheng

Pop Mart

Xpeng

Going Global

Latest

About

Robotics

Collaborate

Earnings

As the global home appliance market moves from rapid growth to tighter competition, innovation in cleaning appliances has slowed. Large household appliances are also under pressure to justify higher prices, forcing companies to rely on substantive technical advances rather than minor feature updates.
That shift set the backdrop for <a href="/company/dreame">Dreame’s</a> media and investor open day on December 17 and 18. Executives overseeing the company’s full product range, from vacuum cleaners to refrigerators, washing machines, air conditioners, small kitchen appliances, and water purifiers, outlined how Dreame is moving beyond single-purpose cleaning products toward a broader smart home lineup.
In the vacuum cleaner segment, previous growth cycles were largely driven by generational improvements in design and core performance. That model is losing momentum. Many brands, executives said, remain locked in marginal and largely homogeneous upgrades that fail to address deeper user pain points.
Since early this year, Dreame has introduced several new components across its vacuum range. One focus is the motor, which plays a direct role in suction and endurance. Dreame said it has developed a digital motor that can reach 200,000 revolutions per minute, which it claims is currently the fastest available. According to the company, the motor improves dust removal from gaps and carpets while allowing longer cleaning sessions without losing power.
To address hair tangling, a persistent industry challenge, Dreame’s floor cleaners use dual blades that the company describes as aviation-grade and titanium-coated. The system reportedly performs up to 2,000 cuts per minute. For wet cleaning, the company has also developed a compact air and water separation module that enables gas and liquid separation within limited space, reducing the traditional tradeoff between suction power and simultaneous wet-dry performance.
Dreame said the upgrades have translated into market gains. Yuan Guangyun, head of R&amp;D at Dreame’s vacuum division, said domestic vacuum cleaner revenue rose 40% year-on-year between January and November this year, while overseas sales increased 59% over the same period.
Beyond cleaning products, two broader shifts are shaping the appliance industry. Consumers are showing greater interest in higher-end products, paying more attention to design, ease of use, and software-driven features. At the same time, appliances are increasingly expected to work together rather than operate as isolated devices.
Drawing on its work in motion control, sensors, and fluid systems, Dreame is applying technologies developed for cleaning products to larger home appliances:
<ul>
<li aria-level="1">In air conditioning, the company said it addressed uneven airflow by launching the X-Wind model, which uses two bionic robotic arms to direct air. The system allows users to control airflow to different parts of a room rather than cooling the entire space evenly.</li>
<li aria-level="1">Refrigerators are another area of focus. Dreame showcased preservation features it said can extend the freshness of meat by up to ten days, as well as light-based temperature control intended to help fruits and vegetables retain moisture. Executives said future models are expected to analyze food conditions and offer dietary suggestions, although those features are still under development.</li>
<li aria-level="1">In small kitchen appliances, Dreame is targeting more flexible cooking. It introduced the DS50 air fryer and the AF60 steam and air fryer, which the company said are designed to cook food more evenly and handle thicker cuts. Along with its kitchen machine and automatic coffee maker, the products broaden Dreame’s presence in the kitchen.</li>
<li aria-level="1">Washing machines and water dispensers are positioned around health and convenience. Dreame said its washing machines can detect fabric types and adjust settings automatically, including low-temperature sterilization. Its water systems combine heating, filtration, and ice-making functions in a single unit.</li>
</ul>
Dreame said its system-level approach is intended to make whole-home intelligence more responsive to comfort and health requirements.
As of June 30, R&amp;D staff made up about 60% of Dreame’s workforce, and R&amp;D spending accounted for at least 7% of the company’s revenue.
Dreame is also expanding production. Construction has begun on its Nanjing manufacturing base, a RMB 4 billion (USD 560 million) project covering about 128,700 square meters. The site will include six production lines for refrigerators and washing machines and is expected to produce close to five million units a year.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3602307154347264">article</a> was written by Huang Nan for 36Kr.

Strong vacuum sales drive Dreame’s push into a broader smart home lineup

East Asia

Dreame

China

Consumer Economy

… strength lies in the “brain,” which represents models, navigation, and operation capabilities.
Previously, SenseTime specialized in large-scale software systems but had no standardized edge products. Through prior investments in hardware and component makers, ACE Robotics now follows an ecosystem model. We define design standards, co-develop hardware with partners, and keep our model layer open, offering base models and training resources.
<h3>36Kr: SenseTime has deep roots in security and autonomous driving. Which of those capabilities translate to robotics?</h3>
WX: R&amp;D systems and safety standards are two key areas. Both autonomous driving and robotics rely on massive datasets for continuous improvement. The validated “data flywheels” we’ve built significantly boost iteration speed. Meanwhile, the rigorous safety and data-quality frameworks from autonomous driving can directly enhance robotics reliability.
On the functional side, our SenseFoundry platform already includes hundreds of modules originally built for fixed-camera city management. When linked to mobile robots, these capabilities transfer seamlessly, extending from static monitoring to dynamic mobility.
<h3>36Kr: How do you view SenseTime’s evolution from visual AI to embodied intelligence?</h3>
WX: SenseTime’s path traces AI’s own progression from version 1.0 to 3.0.
In 2014, we were in the AI 1.0 era, defined by visual recognition. Machines began to outperform the human eye, but intelligence came from manual labeling: tagging images to simulate cognition. Because labeled data was limited and task-specific, each application required its own dataset. Intelligence was only as strong as the amount of human labor behind it. Models were small and lacked generalization across scenarios.
Then came the 2.0 era of large models, which transformed everything. The key difference was data richness. The internet’s texts, poems, and code embody centuries of human knowledge, far more diverse than labeled images. Large models learned from this collective intelligence, allowing them to generalize across industries and domains.
But as online data becomes saturated, the marginal gains from this approach are slowing.
We are now entering the AI 3.0 era of embodied intelligence, defined by direct interaction with the physical world. To truly understand physics and human behavior, reading text and images is no longer enough. AI must engage with the world. Tasks such as cleaning a room or delivering a package demand real-time, adaptive intelligence. Through direct interaction, AI can overcome the limits of existing data and open new pathways for growth.
<h3>36Kr: How does ACE Robotics’ Kairos 3.0 model differ from systems like OpenAI’s Sora or World Labs’ Marble?</h3>
WX: Kairos 3.0 consists of three components: multimodal understanding and fusion, a synthetic network, and behavioral prediction.
The first fuses diverse inputs, including not just images, videos, and text, but also camera poses, 3D object trajectories, and tactile or force data. This enables the model to grasp the real-world physics behind movement and interaction.
In collaboration with Nanyang Technological University, for example, the model can infer camera poses from a single image. When a robotic arm captures a frame, the model can deduce the arm’s position and predict its motion from visual changes, deepening its understanding of physical interaction.
The second component, the synthetic network, can generate videos of robots performing various manipulation tasks, swapping robot types, or altering environmental elements such as objects, tools, or room layouts.
The third, behavioral prediction, enables the model to anticipate a robot’s next move after receiving an instruction, bridging cognition and execution into a complete loop from understanding to action.
<h3>36Kr: How does the human-centric approach improve data efficiency, generalization, and multimodal integration?</h3>
WX: It combines environmental data collection with world modeling.
By “environment,” we mean real human living and working spaces. Unlike autonomous driving, which focuses narrowly on roads, or underwater robotics, we model how humans interact with their surroundings. This yields higher data efficiency and more authentic inputs.
We also integrate human ergonomics, touch, and force, which are all essential for rapid learning, and all missing in machine-centric paths.
<h3>36Kr: When do you expect human-centric systems to achieve large-scale adoption?</h3>
WX: The first large-scale applications will emerge in quadruped robots, or robotic dogs.
Most current quadruped bots still rely on remote control or preset routes. Our system gives them autonomous navigation and spatial intelligence. Equipped with ACE Robotics’ navigation technology, they can coordinate through a control platform, follow Baidu Maps commands, and respond to multimodal or voice inputs. They can identify people in need, record license plates, and detect anomalies.
Linked with our SenseFoundry vision platform, these robots can recognize fights, garbage accumulation, unleashed pets, or unauthorized drones, sending real-time data back to control centers.
This combination, supported by cloud-based management, will soon scale in inspection and monitoring. Within one to two years, we expect widespread deployment in industrial environments.
<h3>36Kr: Beyond the near term, which other applications are worth watching?</h3>
WX: In the medium term, warehouse logistics will likely be the next major commercialization frontier.
Unlike factories, warehouses share consistent operational patterns. As online shopping expands, front-end logistics hubs require standardized automation for sorting and packaging. Traditional robot data collection cannot handle the enormous variety of SKUs, but large-scale environmental data allows our models to generalize and scale efficiently.
In the long term, home environments will be the next key direction, though safety remains a major challenge. Household robots must manage collisions and ensure object safety, much like autonomous driving must evolve from Level 2 to Level 4 autonomy.
Progress is being made. Figure AI, for instance, is partnering with real estate funds managing millions of apartment layouts to gather environmental data, gradually moving embodied intelligence closer to the home.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3594866063835145">article</a> was written by Huang Nan for 36Kr.

SenseTime deepens its push into embodied intelligence with ACE Robotics

SenseTime

Artificial Intelligence

Deep Tech

Embodied Intelligence

… According to Liao, industry capabilities have evolved through five stages:
<ol>
<li aria-level="1">Scripted autonomy, in which robots follow fixed routes within closed environments.</li>
<li aria-level="1">Map-based navigation, which guides robots using high-precision maps to operate along public routes.</li>
<li aria-level="1">Map-light adaptation, enabling real-time adjustments without relying strictly on high-definition maps.</li>
<li aria-level="1">Physically deployable AI agents, which operate autonomously out of the box, planning routes and tasks without degradation.</li>
<li aria-level="1">Cloud and edge integration, enabling coordination among multiple embodied robots for optimized resource allocation.</li>
</ol>
Liao said CowaRobot is currently at the fourth stage and is progressing steadily toward the fifth.
Its system is built around a bird’s-eye view (BEV) world model that predicts uncertain future states and generates actions directly, mimicking human intuition. “When the wind blows trash, the robot anticipates where it will drift,” Liao said. “When it’s too close to a wall, it knows the consequence of collision.”
To handle complex urban features such as traffic lights, no-parking zones, and tactile paving, CowaRobot integrates a visual language model (VLM) as a cognitive layer alongside the physical model. “Think of the VLM as the robot’s brain,” Liao said. “It reasons deeply when needed, then guides motion.”
Through reinforcement learning, this architecture enables training in simulated environments, covering both rare edge cases and coordinated multi-action behaviors.
To make its systems plug-and-play, CowaRobot added two key features. One is a self-memory mechanism that allows robots to recall and adapt to environments over time, improving performance with each pass. The other is prompt-based behavior adjustment, which enables adaptation to local conditions without retraining.
<h3>A decade of data, hardware, and insight</h3>
In embodied intelligence, where deep engineering and long-term commitment are crucial, it is said that time builds moats rather than pressure.
He Tao, CowaRobot’s founder and CEO, holds degrees from Shanghai Jiao Tong University and the Tokyo Institute of Technology, where he also taught. Liao’s background spans control theory and AI, and much of the team shares similar technical depth.
A decade ago, when many companies were racing to deploy robotaxis, CowaRobot took a different route, focusing on perfecting robots in real urban settings first. Over ten years, it has built advantages across three dimensions:
<ol>
<li aria-level="1">Hardware control: From design to manufacturing, CowaRobot develops its own systems, similar to Unitree Robotics’ approach to motor and motion control.</li>
<li aria-level="1">Data scale: The company has reportedly collected 50 petabytes of real-world driving data across sidewalks, parks, and auxiliary roads, which are areas rarely covered in passenger vehicle datasets. It also employs self-supervised learning and VLM-assisted automatic labeling.</li>
<li aria-level="1">Operational insight: A decade of field experience has shaped its understanding of task logic, including when to sweep or spray, when to break minor traffic rules for efficiency, and how to identify illegal roadside waste.</li>
</ol>
<h3>A broader market and a bigger opportunity</h3>
The clearest test of technological maturity is whether the commercial loop closes. Can the system replace human labor, scale across cities, and maintain efficiency over time?
By investing heavily in R&amp;D, CowaRobot believes it has achieved a balance between cost, performance, and adaptability.
Under ideal conditions, one sanitation robot can cover 20–30 kilometers per day, equivalent to the workload of five to ten human cleaners. Even at a modest labor benchmark of RMB 30,000 (USD 4,200) per worker annually, each robot yields a positive gross margin.
Hardware costs have fallen by more than 70 percent from the first-generation model, aided by vertical integration from chassis to cleaning modules. The company also replaced expensive LiDAR (light detection and ranging) sensors with vision-based systems, improving cost-efficiency while maintaining autonomy.
In short, CowaRobot’s advantage lies not only in cost-effectiveness but in AI-driven productivity.
Today, its robots operate routinely across China and are expanding into Singapore and the Middle East. Future applications include litter collection in greenbelts, graffiti removal, and trash can maintenance.
Its new R0 hybrid robot, equipped with dual arms, is already undergoing object pickup tests and is expected to enter the property services and consumer markets soon.
Sanitation automation is far from a niche market. China’s Ministry of Finance reported that in 2024, urban and rural environmental sanitation expenditures reached RMB 242.6 billion (USD 34 billion), with continued growth expected.
As urbanization accelerates, populations age, and labor costs rise, cities will increasingly rely on scalable, intelligent systems to maintain hygiene standards.
With robust engineering talent, complex urban environments suited for training, and vast domestic demand, CowaRobot is well-positioned to help set global benchmarks for embodied intelligence.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3582062216969093">article</a> was written by Xiao Xi for 36Kr.

CowaRobot finds its first real market for embodied robots in sanitation

<a href="/company/baidu">Baidu</a> is rolling out trials of its Apollo Go robotaxis in Switzerland, with further tests planned in the UK and Germany next year, as the Chinese tech giant looks to take on its US peer Waymo in Europe.
Apollo Go has expanded its global reach after trialing robotaxis in Dubai in March, but this will be the first time its vehicles are on European roads.
The test runs in Switzerland, in partnership with PostBus, will use a pilot fleet for mapping trips without passengers and with safety drivers on board.
The company hopes to open its doors to a "select group of users" in the first half of 2026, still with a safety driver onboard, before driverless trials kick off later in the year. It expects to be fully operational in the first quarter of 2027.
The trials come after Baidu announced an agreement in August with ride-hailing platform Lyft to supply self-driving cars in Germany and the UK next year. The European fleet is expected to grow to thousands of vehicles in the following years.
"We are working closely with all relevant stakeholders, including our local partners in the UK and Germany, to advance our plans for market entry and ensure full compliance with local safety and regulatory requirements," Zhang Liang, managing director of Europe, the Middle East and Asia at Baidu Apollo, told Nikkei Asia, although he declined to reveal the timeline.
In October, a few months after Baidu's announcement, US rival Waymo—backed by Google parent Alphabet—also said it would trial robotaxis in London next year with a "small fleet of vehicles," and will look to grow over time.
"We intend to welcome our first public riders next year, contingent on government approvals, and look forward to expanding to more cities around the world in the future," a Waymo spokesperson said.
This makes the British capital the first city where US and Chinese robotaxis will be battling it out. The two giants will also face competition from domestic company Wayve, which is planning to launch pilot trips in London from the spring in partnership with Uber.
Robotaxis have been much slower to take off in Europe than in either the US or China, but BloombergNEF forecasts that there could be as many as 17,000 across Europe by 2030, mostly in the UK and Germany, which are each seen to have over 6,000.
The European market is expected to grow at a compound annual rate of 72.7% to reach USD 8.6 billion by the end of the decade, according to <a href="https://www.grandviewresearch.com/horizon/outlook/robotaxi-market/europe">Grand View Research</a>.
But analysts note that historic European cities could pose a novel set of challenges for Apollo Go and Waymo.
"Street patterns, narrower roads, and, in the case of the UK, driving on the other side of the road to most of the robotaxi development are all challenges for robotaxi developers, but they are not insurmountable," said Andrew Grant, head of intelligent mobility at BloombergNEF, which provides research about clean energy and technologies.
Baidu Apollo's Zhang said he was confident that Apollo Go's experience around the world would enable it to adjust to the specific characteristics of European cities.
"In particular, our expanding test operations in Hong Kong, one of the world's most complex right-hand drive cities, will provide valuable real-world experience that supports a smoother entry into the UK market," he said.
Waymo also said London's winding streets would not pose too much of a problem to its technology. "We are confident in our technology's ability to generalize to new environments like London," the company's spokesperson said.
Robotaxis could face greater competition from public transport in Europe, where many countries have prioritized the development of robobuses over driverless taxis, encouraging some companies to adjust their focus.
<a href="/company/weride">WeRide</a>, another Chinese autonomous vehicle developer, launched new robobus trials in Belgium in September, building off successful runs in France, Spain, and Switzerland.
"Robobuses face lower technical barriers and fit more easily into European transit, especially in dense urban corridors and short 'last-mile' routes connecting transit hubs to final destinations," analysts at S&amp;P said in a recent report.
The looming European competition between Apollo Go and Waymo marks another front in the struggle for technological supremacy between the US and China.
Last month, Apollo Go revealed that the average number of weekly riders surpassed 250,000 in October, a milestone Waymo reached in April.
Waymo has the most fully operational robotaxi services globally, said BloombergNEF's Grant, but all of them are in the US "2026 could be when we see fully operational robotaxi services outside the US and China," he said.
This article first appeared on <a href="https://asia.nikkei.com/business/transportation/uk-europe-emerge-as-robotaxi-battleground-for-baidu-and-waymo">Nikkei Asia</a>. It has been republished here as part of 36Kr’s ongoing partnership with <a href="https://www.kr-asia.com/nikkei-teams-up-with-36kr-in-asia-tech-news-coverage/">Nikkei</a>.

UK, Europe emerge as robotaxi battleground for Baidu and Waymo

Baidu

Transportation

Autonomous Vehicles

Mobility

Europe

… December 4, StairMed unveiled its second-generation high-throughput wireless invasive BCI system. The new version increases the electrode channel count to 256 and expands applications beyond motor control to include speech reconstruction.
<h3>36Kr: StairMed seems to place great emphasis on single-cell brain datasets. Why? Can this be seen as a key part of a data flywheel for brain science?</h3>
ZZ: Exactly. We often describe the BCI ecosystem as an inverted pyramid. At the bottom are the neural interfaces, above that is full-system development, then clinical accessibility, and at the top is neuroscience, which represents the understanding of the brain itself. The degree to which we understand the brain determines how far we can develop BCIs.
The foundational interface system makes clinical data collection possible, and these single-cell datasets are the key to advancing cognitive science. In turn, that deeper understanding helps us design better products and discover new applications.
For any BCI platform, the company that accumulates the largest and richest brain data will be best positioned to define the next wave of applications and technologies.
For example, the long-running US BCI project BrainGate has implanted systems in about 40 to 50 patients, generating limited but highly valuable data. That data has already powered research breakthroughs in thought-based typing and speech decoding. Now imagine scaling that from dozens to thousands or even tens of thousands of implantations.
<h3>36Kr: What role do AI and algorithms play in this process?</h3>
ZZ: Once a large dataset is accumulated for a specific brain region and task, we can train a foundation model using those brain signals, similar to how large language models are trained. This model can then be applied to new patients, giving them stronger baseline decoding and control performance.
Neural network algorithms are well-suited for processing complex data. At the moment, the control signals we decode are relatively simple, say, a two-dimensional cursor movement or a few degrees of motion control. That means smaller models, such as recurrent neural networks with around 100,000 parameters, are sufficient.
In the future, as we scale to higher throughput and more complex controls, both model complexity and parameter counts will rise. I believe AI’s current progress is more than sufficient to meet BCI needs for now.
AI is also valuable on the application side. For instance, in language decoding, we don’t need to decode entire words. If we can extract a few dozen morpheme categories from neural signals, large language models can use context to predict intended words, greatly improving communication efficiency.
Ultimately, the relationship between BCIs, humans, and agentic AI will be dynamic and collaborative. Sometimes we will issue high-level commands through brain signals and let the AI agent execute them, while in other cases, we will fine-tune actions ourselves for greater precision.
<h3>36Kr: Based on the performance of existing implant recipients, do you think BCIs could eventually give humans capabilities beyond the ordinary?</h3>
LX: That's actually what motivated us to pursue this technology in the first place. We’ve always envisioned consumer-grade applications. We started with medical use because it's a necessary path: it meets urgent clinical needs and forces us to perfect the technology.
Our hope is not just to restore motor or sensory function, but also to push the boundaries of human control. In the past, we controlled limbs with our brains. In the future, we might control external devices via BCIs, potentially surpassing normal human limits.
For example, Neuralink once released a mini game to measure cursor control speed and accuracy. Ordinary users can reach eight to ten bits per second (BPS), but paralyzed patients using BCIs achieved 9.5 BPS after training, which is faster than most healthy users. That shows BCIs can already confer comparative advantages in specific tasks.
Another idea: today, all device control remains constrained by path dependency. To move a cursor from one corner of the screen to another, you must physically drag it. In the future, by decoding spatial memory signals from the hippocampus, we could “teleport” the cursor instead, which is impossible with current technology.
Similarly, conventional controllers or voice commands are crude tools for manipulating robotic arms with multiple degrees of freedom. BCIs, however, can encode far more nuanced neural information, allowing users to control robotic limbs as naturally as their own. Research in leading journals has already shown that trained BCI users can operate robotic arms with smooth, coordinated motion.
<h3>36Kr: With AI evolving so rapidly, many experts worry about its impact on human education, employment, and even existence. Do you think human-machine fusion is inevitable?</h3>
LX: AI is meant to serve humans, and human intent should always remain central. But we cannot rule out the possibility that AI may one day develop self-awareness.
To prevent humans from losing agency or relevance, deep integration between humans and machines is, in my view, inevitable. The key to that integration is the bridge, and that is exactly what BCIs represent: a channel for information exchange between humans and machines.
Without that bridge, silicon-based AI and carbon-based humans could drift apart. But if AI becomes part of us, or we become part of it, we could achieve a harmonious coexistence, one without communication barriers.
Think of autonomous driving: you can tell the car to go from point A to point B, and it will plan and navigate independently, yet you can always take the wheel. That is controlled evolution. What we must avoid is a future where AI or robots operate entirely beyond human control. Deep human-machine integration may be the safer, more balanced path forward.
<h3>36Kr: If BCIs become fully mature, would you personally get implanted?</h3>
LX: We've talked about it before. I'd love to try it myself, just to experience what it actually feels like.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3569598798543750">article</a> was written by Hai Ruojing for 36Kr.

How far are brain-computer interfaces from restoring and extending human ability?

Medtech

… added that it expects to be the first in China to mass produce and deliver automotive-grade Level 4 autonomous logistics vehicles.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3597598153785352">article</a> was written by Xu Caiyu for 36Kr.

Rino.ai raises new funding to accelerate toward mass production of autonomous cargo vans

Logistics

<a href="/company/li-auto">Li Auto</a> is setting up an <a href="/industry/ai">artificial intelligence</a> R&amp;D center in Silicon Valley, 36Kr has learned from multiple sources. The center will focus on smart driving technologies, and recruitment has been underway for several months.
According to a person familiar with the matter, Li Auto has long maintained a lean North American R&amp;D team supporting chip development and other intelligent systems. “This time, the plan is to upgrade the Silicon Valley team into a full-fledged R&amp;D center,” the source said.
36Kr understands that the new center will primarily focus on autonomous and assisted driving, with Li Auto seeking senior algorithm engineers with experience in advanced AI research.
Since October, Li Auto has completed channel and product rollouts in four overseas markets, including Egypt and Kazakhstan. Its global expansion strategy places similar emphasis on research. The company already operates R&amp;D centers in Germany and the US.
With the addition of the Silicon Valley facility, Li Auto will run four R&amp;D hubs worldwide. Its first overseas R&amp;D center, in Munich, Germany, opened in January 2024 and focuses on advanced design, power semiconductors, chassis systems, and electric drive technologies. Domestically, the company operates R&amp;D centers in Beijing and Shanghai, which are responsible for core technology development and vehicle engineering.
Home to <a href="/company/tesla">Tesla</a>, Waymo, <a href="/company/openai">OpenAI</a>, and <a href="/company/nvidia">Nvidia</a>, Silicon Valley remains one of the world’s densest concentrations of AI talent. For Chinese automotive companies, a presence in the region offers closer proximity to frontier research and experienced engineers.
Among China’s newer <a href="/industry/ev">electric vehicle</a> manufacturers, <a href="/company/nio">Nio</a> and <a href="/company/xpeng">Xpeng</a> established Silicon Valley operations earlier, Nio in 2014 and Xpeng in 2018. Li Auto’s later entry, with a focus on autonomous driving, reflects intensifying competition among Chinese automakers, as industry executives increasingly point to smart driving as a core area of focus next year.
That competitive pressure is rising across the industry. This year, <a href="/company/huawei">Huawei</a> launched ADS 4.0, its Level 3 autonomous driving solution for highway scenarios. Vehicles including the Maextro S800 and Aito M8 are now equipped with hardware capable of supporting Level 3 functions.
In March, Li Auto introduced its vision-language-action (VLA) model, which jointly models visual, linguistic, and motion data. By incorporating language modeling and reasoning chains, the system is designed to interpret complex driving environments, infer causality, and make decisions with less reliance on rule-based logic, potentially improving performance in unfamiliar scenarios.
On September 10, Li Auto began a broad rollout of the VLA model to users. CEO Li Xiang said on social media that the released version was a lite edition, with several key functions not yet activated.
Integrating large models into autonomous driving systems remains technically challenging. While the potential benefits are clear, reliable real-world performance depends on continued advances in algorithms, computing power, and data, validated through large-scale road testing.
China’s limited pool of engineers with experience spanning both large models and autonomous driving has made overseas hiring increasingly attractive. Against that backdrop, Li Auto’s move into Silicon Valley could prove strategically significant.
whereRelatedArticle
Xpeng, whose core strength lies in assisted driving, has operated R&amp;D centers in North America since 2018. Its teams in Silicon Valley and San Diego together number about 100 people. Even during periods of cost control, when hiring was restricted, the company retained the group.
That North American team has since become central to Xpeng’s driving technology development, handling algorithm work while China-based teams focus on requirements definition and testing.
In September, CEO He Xiaopeng allocated additional computing resources to the North American team to accelerate deployment of the company’s foundation model, which is being developed for in-vehicle use within the year.
The project is led by Liu Xianming, an executive Xpeng recruited from Cruise. According to company disclosures, the cloud-based foundation model has 72 billion parameters, is trained in the cloud, and is designed to transfer knowledge to a smaller model optimized for in-vehicle chips.
“The car-side model will likely have around seven to eight billion parameters, and internal expectations are high,” one person familiar with the project told 36Kr. By comparison, Li Auto’s deployed VLA model operates with roughly four billion parameters.
Xpeng’s sustained investment in North America has extended beyond vehicles. The company also operates a robotics team of about 20 people in the region. When Xpeng unveiled its first humanoid robot, Iron, in November, its share price rose, reflecting investor interest in its broader strategy.
At the start of 2025, Li Xiang outlined an updated corporate vision, saying Li Auto aims to become an “AI company” over the next decade. Business units have since reevaluated their operations through an AI-first framework. The VLA model marks an initial step, while the Silicon Valley center signals a deeper commitment.
Xpeng CEO He has said publicly that multimodal learning, world modeling, and long-horizon planning, capabilities often associated with artificial general intelligence (AGI), remain underdeveloped and may take years to mature or require fundamental breakthroughs.
As the smart vehicle framework continues to evolve, automakers face a narrowing margin for error. The ability to identify, absorb, and deploy new technologies may increasingly determine long-term competitiveness. That dynamic may help explain why China’s leading EV companies are converging on Silicon Valley, positioning themselves closer to one of the world’s most concentrated sources of AI innovation.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3466903693399680">article</a> was written by Xu Caiyu for 36Kr.

Li Auto builds Silicon Valley lab to chase Xpeng and others in smart driving

Electric Vehicles

… Happily Ever After. That’s a rare practice in the industry.
Jia Fei, however, understands how to build lasting IP. In 2016, he created Little Monk Yichan, a viral character that amassed 150 million fans across platforms.
That experience now shapes his approach to developing Liu’s persona. A year ago, Liu appeared for only five minutes in Affectionate Seduction, yet viewers responded strongly to him. Sensing potential, Jia Fei quickly edited a fan video spotlighting Liu’s scenes and posted it online, drawing the first wave of fans.
Over the following year, the team continued to create short skits around the same character archetype—disciplined, restrained, and quietly affectionate—traits that mirrored Liu himself. “He’s a traditional Shandong man,” Jia Fei said. “That self-restraint and integrity just radiate from him.”
So when Summer Rose premiered, fans’ emotions were primed to erupt, creating a perfect storm of connection that turned Liu into the next big star.
His breakout also marks a new chapter for the industry. The days of waiting for luck and miracles are over. Now, short drama makers are trying to engineer virality, to understand how to make the next Zhao Kechun or Liu Xiaoxu.
At the Weibo TV &amp; Internet Video Summit in early November, the red carpet featured major celebrities like Yang Mi and Wang Hedi, alongside short drama sensations such as Zhao, Yu Yin, and Yican Media’s own Wang Gege. Jia Fei splurged on a top-tier makeup artist and photographer, each charging tens of thousands of RMB per day. “Expensive,” he said, “but Gege deserves it.”
Hema Xingchi went the opposite route, branding Liu Boyang as “short drama’s number one bad boyfriend,” playing on his string of “repentant ex-lover” roles. Though Liu was hesitant at first, he agreed, realizing that the bigger risk was blending into the crowd.
Still, shadows linger over these shining stars.
Turnover is brutal. In just over a year, first-generation icons like Xu Yizhen and Sun Yue have faded from the spotlight. On Hongguo, the popularity rankings shift almost every month.
Industry veterans and viewers alike say part of short drama stars’ appeal lies in their relatability, a contrast to the carefully curated personas of traditional celebrities. Yet as commercialization deepens, that down-to-earth charm is eroding.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3586384697703553">article</a> was written by Lan Jie for 36Kr.

How short dramas are turning everyday actors into China’s newest idols

Entertainment

Video

Creative

Content

… user experience, you end up with clunky interactions, and that makes it hard for users to build emotional bonds with your brand,” Tong said.
Whether serving pets or people, Enabot’s mission has stayed consistent: to refine and deepen value in specific scenarios. This philosophy reflects a new generation of hardware companies focused on doing one thing exceptionally well rather than attempting everything at once.
<figure id="attachment_159003" aria-describedby="caption-attachment-159003" style="width: 480px" class="wp-caption aligncenter"><img loading="lazy" decoding="async" class="wp-image-159003" src="https://console.kr-asia.com/wp-content/uploads/2025/12/Enabot-Production-Facility.jpg" alt="Photo shows Enabot’s production facility." width="480" height="480" srcset="https://console.kr-asia.com/wp-content/uploads/2025/12/Enabot-Production-Facility.jpg 1080w, https://console.kr-asia.com/wp-content/uploads/2025/12/Enabot-Production-Facility-720x720.jpg 720w, https://console.kr-asia.com/wp-content/uploads/2025/12/Enabot-Production-Facility-1024x1024.jpg 1024w, https://console.kr-asia.com/wp-content/uploads/2025/12/Enabot-Production-Facility-360x360.jpg 360w, https://console.kr-asia.com/wp-content/uploads/2025/12/Enabot-Production-Facility-768x768.jpg 768w, https://console.kr-asia.com/wp-content/uploads/2025/12/Enabot-Production-Facility-600x600.jpg 600w, https://console.kr-asia.com/wp-content/uploads/2025/12/Enabot-Production-Facility-300x300.jpg 300w" sizes="(max-width: 480px) 100vw, 480px" /><figcaption id="caption-attachment-159003" class="wp-caption-text">Enabot’s production facility. Photo source: Enabot.</figcaption></figure>
<h3>Digital friends</h3>
In the post-pandemic era, the <a href="https://www.mckinsey.com/featured-insights/asia-pacific/five-fifty-the-lonely-economy">“loneliness economy”</a> is growing faster than ever. From young professionals living alone to self-reliant seniors and overworked office employees seeking emotional relief, companion robots are emerging as a new form of connection.
As these robots reach global markets, cultural context increasingly shapes how users perceive and interact with them.
In East Asia, and particularly China, companion robots often serve as emotional bridges within families, extending beyond the role of smart devices.
This reflects regional family structures and emotional norms. Enabot’s research shows that its core users include dual-income families and those with members living apart. Their primary need is for “remote co-presence,” or a sense of emotional connection across distance.
To meet that need, Enabot added features imbued with ritual and warmth. On a family member’s birthday, users can remotely adjust the robot’s eye light to display celebratory images. Its customizable facial expressions allow playful messages or emotive animations, fitting the East Asian preference for frequent, close family interaction.
<figure id="attachment_159002" aria-describedby="caption-attachment-159002" style="width: 1920px" class="wp-caption alignnone"><img loading="lazy" decoding="async" class="size-full wp-image-159002" src="https://console.kr-asia.com/wp-content/uploads/2025/12/Enabot-Companionship.jpg" alt="" width="1920" height="1280" srcset="https://console.kr-asia.com/wp-content/uploads/2025/12/Enabot-Companionship.jpg 1920w, https://console.kr-asia.com/wp-content/uploads/2025/12/Enabot-Companionship-720x480.jpg 720w, https://console.kr-asia.com/wp-content/uploads/2025/12/Enabot-Companionship-1024x683.jpg 1024w, https://console.kr-asia.com/wp-content/uploads/2025/12/Enabot-Companionship-768x512.jpg 768w, https://console.kr-asia.com/wp-content/uploads/2025/12/Enabot-Companionship-1536x1024.jpg 1536w" sizes="(max-width: 1920px) 100vw, 1920px" /><figcaption id="caption-attachment-159002" class="wp-caption-text">Photo source: Enabot.</figcaption></figure>
In contrast, users in Europe and the US prefer robots with humanlike empathy but lower intrusion. “Foreign users tend to want companionship that’s less intrusive,” Tong told 36Kr. “They appreciate a robot’s presence but do not want to be constantly interrupted.”
This preference reflects the Western emphasis on personal space and the independence typical of smaller family structures.
In practice, Enabot’s robots often function as discreet yet capable assistants. For security, they serve as mobile guards that monitor environmental changes and alert users to anomalies, offering an advantage over fixed cameras.
In companionship scenarios, users expect the robots to remember preferences, adapt to personal routines, and collaborate creatively, whether brainstorming story ideas or designing simple interactive games.
Western users also place high value on autonomy. When a device malfunctions, they expect it to self-diagnose and attempt repairs before sending logs to a server, reducing human intervention and safeguarding privacy through closed-loop data handling.
Ultimately, companion robots are the physical embodiment of human relationships. Across markets, they extend family bonds in some cultures and preserve personal boundaries in others. The contrast reveals a shared truth: understanding emotional reality is the foundation of meaningful connection.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3541396662431875">article</a> was written by Huang Nan for 36Kr.

Inside Enabot’s quest to build companion robots first for pets, then for people