Pogbagram

I recently moved into my newly renovated home, marking the transition from a non-smart home era to a smart home era. Before bed, I wanted to set up a scenario where only the bedroom lights would be on and at half brightness, but I discovered that I would have to manually configure twenty or thirty lights throughout the house. I gave up halfway through and ended up getting out of bed to turn them off one by one.

It turns out that smart homes are not just about "connecting devices to the internet and controlling them with an app"—sometimes, they are even more complicated than non-smart homes.

Whole-house networking has been around for many years, but a whole house doesn't necessarily qualify as smart.

The Internet of Things has been talked about for over a decade, yet users still experience a familiar sense of disappointment. Devices are becoming increasingly numerous, connectivity is becoming easier, and voice control is no longer a novelty, but when it comes to the home, the so-called "intelligence" often remains superficial—one command corresponds to one action. The system executes, but it doesn't understand; it can coordinate, but it doesn't act proactively.

Admittedly, it is not easy for home appliances to fully understand human speech. They need to have at least six core capabilities: connection, perception, reasoning, execution, memory, and optimization. However, reasoning and memory are often the weak points of most home appliances.

For example, the instruction to "close all the curtains in a room with the lights on" is easy for a person to understand, but home appliances may only grasp the key words—either turn on the lights or close all the curtains. Similarly, if you usually set the air conditioner to 26°C when you sleep, but say "I'm going to sleep," the air conditioner will only mechanically turn on and won't automatically adjust to 26°C.

The absence of reasoning and memory presents systemic challenges. Xu Yi, Dean of Midea AI Research Institute, points out that the challenges at the reasoning level mainly lie in the inherent heterogeneity of device capabilities, varying home environments, the inherent ambiguity of user input, the individual differences in haptic judgment, and finding a balance between rapid response and accurate understanding. The difficulty at the memory level lies in distinguishing between immediate information and long-term preferences, forming associative structured knowledge, recalling the correct memory at the right time, and dynamically adjusting to changes in user habits. Ultimately, memory needs to be integrated with reasoning to allow experience to be accumulated into long-term capabilities.

To solve these problems, the capabilities of smart homes must be reconstructed from the ground up. Midea's recently released MevoX, the industry's first self-evolving smart home system, was created precisely to address these challenges.

MevoX possesses two core capabilities: advanced reasoning and persistent memory. "Reasoning" goes beyond simply translating a sentence into an instruction; it incorporates information from multiple dimensions, including time, space, state, scenario, and location, to infer the user's true desired outcome. The "memory" capability is divided into short-term, medium-term, and long-term layers. It doesn't memorize rigid commands, but rather user preferences and habits. For example, if a user once says, "I'm afraid of the cold when I sleep," the system will retain this information as a basis for adjusting the nighttime environment.

Overall, MevoX, as an AI brain that can understand users, has achieved a closed loop of "user expressing feelings – system understanding intent – space execution plan", upgrading whole-house intelligence from "human-driven devices" to "intent-driven space", and becoming more user-aware after each interaction.

MIA 1.0: Transforming the Home from "Device Interconnection" to "Autonomous Driving"

With MevoX as its AI brain, a unified scheduling system is still needed to truly enable all devices in the home to work together. To this end, Midea created the Home Intelligent Navigation System (MIA 1.0). While the name sounds somewhat like something from the automotive industry, its purpose is quite apt: if individual appliances are merely parts that perform tasks, then MIA 1.0 is like the central scheduling system for the home space. It's responsible for coordinating device scheduling, making optimal decisions, and enabling the home to move from single-product intelligence to system intelligence, and then to multi-agent collaboration. Midea's own analogy is a "home autonomous driving system." This term is somewhat ambitious, but it's certainly closer to what it aims to achieve than simply "interconnection."

In the logic of MIA 1.0, intelligence is achieved through multiple parallel processes.

The first line is individual product intelligence. Each device first solidifies its own professional capabilities. Air conditioners understand air better, washing machines understand washing and care better, and water purification equipment understands water safety better.

The second line is system intelligence. Air conditioning, fans, air purifiers, humidifiers, and air sensors make up the air system; kitchen, laundry, lighting, security, and energy also form more vertical scene systems. They are no longer simple mechanical linkages, but work continuously around a life goal.

The third theme is multi-agent collaboration, a key point repeatedly mentioned at this press conference. Smart homes are beginning to extend to external terminals such as mobile phones and cars, forming a more complete experience of the flow between people, cars, and homes. Midea is promoting the implementation of cutting-edge technologies such as A2A (Agent to Agent) and has already partnered with several mainstream mobile phone manufacturers and car companies. You can understand it as: intelligent agents in different spaces beginning to collaborate around the same life task, instead of each doing their own thing.

As a result, MIA 1.0 achieves unified scheduling and optimal decision-making for all smart devices in the house, truly enabling the home to have thinking, operating, and service capabilities, realizing the "autonomous driving" experience of the home.

When MevoX endows the whole-house smart home with the cognitive ability of "self-evolution," and MIA 1.0 endows it with the execution capability of "proactive service," the role of the home begins to undergo a fundamental change: it is no longer a toolbox waiting for instructions, but a "living entity" capable of sensing, thinking, and acting. Devices learn to proactively serve, spaces learn to dynamically adapt, and the home truly comes alive.

Midea aims to be both a "home appliance master" and a "smart technology expert."

A very interesting phenomenon exists in the current automotive industry: many established automakers, having focused primarily on hardware, have neglected or even ignored investment in smart cockpits and driver assistance systems. Faced with competition from emerging electric vehicle brands, they appear outdated and obsolete. After much reflection, they have had no choice but to bring in external partners to address these shortcomings.

A similar situation has occurred in the home appliance industry.

In the past, the industry commonly adopted two approaches: one group was more knowledgeable about hardware and home appliances but less adept at AI systems; the other group was very knowledgeable about intelligent interaction and software capabilities, but when it came to specific home appliances, spaces, and families, they were prone to the problem of "talking the talk but not walking the walk." Midea, however, is truly integrating the two—being both a "home appliance master" and a "smart expert."

Midea has 500 million home appliances across all categories with connectivity capabilities, more than 140 million smart home appliances connected globally, and more than 150 million smart users connected, completing the AI-enabled deployment of more than 150 categories of home appliances.

Many AI applications are struggling with issues like acquiring new users, user activity, and user retention. For Midea, however, all they need to consider is using AI to revitalize their hundreds of millions of products and users.

This is similar to the logic of "giving civilization to time, rather than giving time to civilization," and Midea needs to give AI to its products.

In an interview, Xu Yi also provided an important piece of information when discussing the group's investment in AI:

The AI ​​Research Institute will continue to support all of the group's businesses. Furthermore, the AI ​​Research Institute has been continuously recruiting and investing, including in computing power and infrastructure construction. Midea has also previously announced that it will invest 60 billion yuan in AI and cutting-edge research over the next three years.

Midea AI Research Institute was established in 2020, two years before the world-changing ChatGPT 3.5 was released. However, both in terms of the timing of its establishment and the scale of its investment, Midea's attitude towards AI is the most forward-looking in the industry. Moreover, Midea's AI strategy has long been limited to home appliances and furniture, but has expanded to fields such as smart buildings, embody intelligence, and energy.

Those who go it alone go fast, but those who go together go far. Midea has also explicitly proposed building "a more diverse intelligent open platform," aiming to create a win-win situation for users through win-win cooperation with five major ecosystem partners: mobile phones, automobiles, content, hardware, and AI. Currently, Midea has already partnered with mobile phone brands such as Honor, vivo, and OPPO, as well as automakers such as BYD, NIO, and Changan.

This series of actions is making the industry re-understand the essence of whole-house intelligence: it is not isolated islands, but an open and win-win ecosystem; it is not a high-level display of technology, but a people-oriented service.

Specifically, at the level of people and homes, the significance of MevoX is actually quite simple. It allows people to gradually shift from adapting to devices to allowing devices to adapt to people. You don't need to remember so many commands, repeatedly explain yourself, or start setting up your life from scratch every time. Your home can remember you, infer your needs, and handle those small but frequent decisions for you at the right time. When should the lights turn on, when should the air conditioning be adjusted, when should the laundry and skincare products be reminded, when should the kitchen be involved? The system begins to take on these trivial judgments. Life doesn't become more futuristic as a result, but rather, it finally reduces the fatigue of being managed by devices.

As Xu Yi said, MevoX's mission is to transform smart home appliances from "tools" into "family members." Thus, the meaning of home is redefined; it is no longer a house that gets older and older, but a space that gets newer, more worry-free, and happier with each passing year.

#Welcome to follow iFanr's official WeChat account: iFanr (WeChat ID: ifanr), where more exciting content will be presented to you as soon as possible.

Last week, a mysterious model began circulating in the lobster community: Pony-Alpha-2.

Without a press conference or official introduction, it quietly ran within Zhipu's AutoClaw platform, open to user beta testing. Unexpectedly, word-of-mouth spread rapidly, leading everyone to ask: What exactly is this anonymous AI model?

Today, the mystery has been solved. It is GLM-5-Turbo, the "world's first lobster model" officially released by Zhipu today, and also Zhipu's first closed-source model since 2025.

Why does the general-purpose model keep failing when running lobsters?

Anyone who's ever raised a lobster probably knows this: making the model chat or write goes smoothly. But when it comes to actually running tasks, problems often start to appear around the third step. Tool calls fail, context is forgotten, long tasks crash midway—all sorts of things happen.

The root of the problem lies not in the framework, but in the underlying model itself. The training objective of a general-purpose large model is dialogue, while executing workflows is a completely different matter. Once it enters a real-world lobster scenario with multiple steps, multiple tools linked together, and continuous execution, it falls short.

GLM-5-Turbo was designed to solve this problem. From training data construction to optimization objectives, the entire process was specifically optimized for the lobster scenario, focusing on training five key capabilities.

In terms of tool invocation, stable invocation of external tools and various skills is enhanced, and uninterrupted operation is a basic requirement. Regarding instruction compliance, complex and multi-layered long instructions can be accurately broken down, supporting target identification, step planning, and multi-agent collaboration.

Scheduled and continuous tasks are another key focus. The model can understand instructions over time, handle time-triggered scenarios, and ensure uninterrupted execution of long tasks. Its programming capabilities continue the strengths of the GLM series, enabling long-range agentic engineering tasks with minimal human intervention, moving from Vibe Coding to Agentic Engineering. In terms of speed, it has also made specific speed improvements for high-throughput, long-chain scenarios, and its response stability leads similar models.

In terms of evaluation, GLM-5-Turbo achieved the top ranking among domestic models in ZClawBench, the end-to-end benchmark for lobster scenarios developed by Zhipu.

The creation of ZClawBench has a specific background.

With the increasing popularity of OpenClaw, the current task types now cover a variety of scenarios such as installation and configuration, code development, information gathering, data analysis, and content creation.

The user base has expanded from early developers to include efficiency-conscious office workers, financial professionals, operations engineers, content creators, and research analysts. ZClawBench is built upon these real-world use cases, and its question bank and test tracks are now fully publicly available.

In addition, the GLM-5-Turbo is also integrated with the Mechrevo box from iSoftStone, creating a native AI Agent terminal experience. However, there's a side note worth mentioning: the API price of the GLM-5-Turbo has also seen its second price increase this year, rising by 20% compared to the GLM-5.

Although considering that lobster-related tasks often involve dozens of tool calls and contextual connections, the token consumption for a typical cross-departmental meeting minutes compilation and task distribution far exceeds that of ordinary conversations, the dimensions of cost accounting for enterprises are also changing—

The amount of tokens spent is no longer the key point; the most important thing is how much manpower the AI ​​employees save.

Along with the model, lobster combo meals for individuals and businesses are also being released, with both individual and team versions available.

Whether you're a one-person company, a startup team, or a large or medium-sized enterprise, you can flexibly subscribe based on your business scale, with a maximum of 5 tokens per account. Enterprises can subscribe flexibly based on the actual number of employees through the Team version, ensuring token supply and stable online AI employees at a controllable cost.

The accompanying enterprise-level security management system, "Claw for Enterprise Security," supports unified scheduling and permission orchestration for different types of Claws. Through a centralized management console, the execution path, tool call chain, and resource consumption of each agent task can be monitored and visualized in real time.

Lobster model, targeting enterprise entry point

Reading the introduction alone isn't enough; I tested it with a real task and here's my experience.

The first scenario is very simple: set an alarm for 10 minutes later, and when the time comes, Lark will remind you to "get up and move around." The model directly replies that the 14:22 reminder has been set, and Lark receives the message precisely 10 minutes later, without any steps requiring manual intervention.

The second scenario involved information gathering, summarizing the day's hot topics in the tech world. The model was run through a testing tool, compiling key information such as Nvidia's GTC conference, Musk's announcement of the Terafab chip-making project starting in seven days, and the addition of six new key future industries by the government.

The last scenario, which is just for fun, is a bit more complicated. The lobster runs on a cloud computer provided by Lark, which is an Alibaba Cloud ECS virtual machine. I had it write a temperature monitoring skill, encapsulate it, and automatically load and activate it, issuing a Lark alert if the temperature exceeds 40 degrees Celsius. The problem is, cloud virtual machines don't have physical temperature sensors, so directly reading the temperature won't work.

The model didn't get stuck waiting for me to give it directions. Instead, it tried five different reading methods in turn. After none of them worked, it switched to using CPU load as a proxy indicator to estimate the temperature and told me why it did this.

After writing the script and getting it running locally, I wrote SKILL.md, registered it in openclaw.json, triggered a restart, and ran doctor to confirm successful loading. Finally, Lark sent a confirmation: the current temperature is 27°C, and everything is normal. The entire process was uninterrupted and didn't wait for me to send the next command.

After completing the three tasks, the tool performed flawlessly, the timing instructions were accurately interpreted, and the information gathered yielded conclusions that went beyond mere listing. Of course, these are relatively standard lobster-themed tasks; more complex multi-agent collaborative scenarios require further real-world testing.

As of press time, GLM-5-Turbo is now online. Developers and enterprise users can call the API through the BigModel.cn or api.z.ai open platform, or access it through the Lobster Package.

GLM Coding Plan Max has been integrated into GLM-5-Turbo. The Pro plan will support it this month, and the Lite plan will be available in April. For agent scenarios requiring long-term stable operation, continuous background execution, or higher load, the Lobster plan is recommended.

The default model for AutoClaw Aolong has been switched to GLM-5-Turbo. From now until March 22, there is a limited-time discount on fuel packages, with a maximum discount of 34%. The fuel package is valid for 7 days from the date of purchase.

It's worth noting that the lobster scenario allowed ordinary users to truly experience the difference between "AI doing chores for me" and "AI answering my questions." The former requires completely different capabilities, which explains the significant gap between models specifically trained for lobster scenarios and general-purpose models.

In the agent era, competition may ultimately not be about the level of general capabilities, but rather who can truly operate smoothly, stably, and be used by enterprises as a productivity tool in specific scenarios. Capabilities are merely the entry ticket; the real test is the ability to consistently deliver results.

Here is the link to try it out:

• AutoClaw Aolong Client: https://autoglm.zhipuai.cn/autoclaw
• Z.ai：z.ai
• Zhipu Qingyan APP/Web Version

#Welcome to follow iFanr's official WeChat account: iFanr (WeChat ID: ifanr), where more exciting content will be presented to you as soon as possible.

In the past few years, the idea that "joint ventures are falling behind" has become an unspoken consensus in the industry.

In core dimensions such as intelligence, range, and cabin space, joint venture brands are facing severe challenges from emerging domestic brands. The brand momentum they accumulated in the era of gasoline vehicles is being rapidly depleted in the wave of new energy vehicles.

Even SAIC Volkswagen, which once dominated sales during the era of gasoline-powered vehicles, is no exception. Although its core market share remains, its influence in the public sphere has diminished significantly.

Faced with this predicament, Volkswagen has had to try adopting the strategies of emerging electric vehicle manufacturers. On the afternoon of the 16th, SAIC Volkswagen held a technology launch event for its upcoming six-seat extended-range SUV, the ID. ERA 9X.

When asked about the past controversy with Li Auto regarding range-extending technology, SAIC Volkswagen executive Fu Qiang frankly admitted the misjudgment at the time and demonstrated Volkswagen's determination to correct its mistakes and seek change with the phrase "thank you for the efforts of Chinese automakers".

Therefore, the launch of the ID. ERA 9X carries significance beyond the product itself. It's not just another SUV that Volkswagen has launched in China; it sends a strong signal: this German automaker is finally starting to build cars with Chinese speed and Chinese thinking.

China's Solution

The appearance is the first thing the ID. ERA 9X presents to the market.

The vehicle closely resembles the design of the concept car, with a square and upright silhouette. The closed-off front fascia, combined with the continuous LED light strip and illuminated logo, exudes a futuristic and technological feel; the two rectangular LED daytime running lights below the bumper further enhance the front's recognizability.

Moving to the side profile, the designers cleverly created a floating roof effect by blacking out all the pillars, retaining the imposing presence of a flagship SUV while injecting a touch of lightness into the body lines. The power-closing doors, illuminated trim pieces on the front and rear bumpers, and multi-spoke wheels up to 21 inches perfectly balance luxury and sportiness.

In terms of dimensions, the new vehicle measures 5207/1997/1810mm in length, width, and height, respectively, with a wheelbase of 3070mm. This larger size surpasses that of the Range Rover, making it the largest SUV currently sold globally by the Volkswagen brand.

In addition, Volkswagen offers six exterior colors for the new car: Fairy Green, Desert Gold, Sky Blue, Snow White, Distant Mountain Gray, and Obsidian Black, while the interior offers three themes: Natural (Beige), Brocade Orange, and Ink Treasure.

If the imposing exterior is the key to unlocking the door, then the intelligent cockpit of the ID. ERA 9X is the core answer that truly demonstrates its "sincerity in localization".

Thanks to its extra-long wheelbase of 3070mm and boxy body structure, the second and third rows of the vehicle offer extremely ample headroom and legroom. The third row has completely eliminated the awkwardness of being just an "emergency bench," truly meeting the needs of a full family for high-quality travel.

In terms of cabin configuration, the new car features a 15.6-inch dual-screen setup in the front, integrating the driver's instrument panel and central control functions; the rear seats are equipped with a 21.4-inch foldable ceiling-mounted screen.

All screens in the vehicle feature viewing angle compensation films, thoughtfully catering to the viewing experience of passengers in different positions. Particularly noteworthy is the innovative integration of the Smart Surface Magic Screen into the interior wood trim. This design, seamlessly switching between a hidden state and touch operation, is truly unique among currently mass-produced vehicles in China.

Furthermore, from the AR-HUD head-up display, second-row zero-gravity seats, and intelligent refrigerator, to the intelligent voice system that supports independent wake-up in four sound zones, 5G vehicle networking, and active noise cancellation technology, the ID. ERA 9X's cabin configuration has fully aligned with domestic high-end new energy vehicles, completely making up for the shortcomings of joint venture brands in terms of comfort and technology.

In the field of intelligent driving, which is increasingly valued by consumers, Volkswagen decisively chose the mature "China solution".

The new vehicle is equipped with an advanced intelligent driving system deeply powered by Momenta. With the support of powerful hardware such as roof-mounted lidar, millimeter-wave radar, and high-definition cameras, the new vehicle can achieve map-free NOA (No-Action) navigation assistance covering all scenarios, including highways, urban areas, and point-to-point driving. Whether it's automatic following, intelligent lane changing, or intelligent parking and door opening warning in complex scenarios, its intelligent driving performance has already entered the top tier of the industry.

German heritage

The powertrain and chassis tuning of the ID. ERA 9X are the best proof of Volkswagen's perfect integration of its profound German engineering heritage with the needs of the Chinese market.

The new vehicle is equipped with the EA211 EVO II series 105kW engine, which was developed specifically for range-extended vehicles. The company confidently calls it "the most powerful range extender on earth".

By integrating advanced technologies such as Miller cycle, VTG variable geometry turbine, and 350bar high-pressure direct injection, this range extender experiences only about 5% power loss under low-battery conditions (compared to 50% to 150% for competing products in the same class).

Even in extremely cold environments of -30°C, its 0-100 km/h acceleration time can still reach 6.31 seconds, and the cabin noise increase when the range extender intervenes is less than 0.5 dB, truly achieving imperceptible intervention.

The new vehicle is currently available in two powertrain options: a single-motor rear-wheel-drive version with a maximum power of 220kW, and a dual-motor four-wheel-drive version with a combined power of 380kW (160kW front / 220kW rear).

In terms of batteries, Volkswagen offers two options: a 51.1kWh lithium iron phosphate battery and a 65.2kWh CATL ternary lithium battery. Under CLTC conditions, the maximum pure electric range exceeds 400km, and the combined range exceeds 1000km.

To allow this 5.2-meter-long behemoth to navigate the city with ease, Volkswagen equipped it with a chassis system that surpasses its class.

The combination of a double wishbone and a five-link independent suspension at the rear, along with optional dual-chamber air springs and DCC continuously adjustable dampers, achieves a perfect balance between comfortable vibration filtering and the stable driving feel of German engineering.

Even more noteworthy is that, thanks to the addition of rear-wheel steering technology, the ID. ERA 9X's minimum turning diameter has been significantly reduced to 11.2 meters, making this large SUV as maneuverable in tight parking spaces as a compact sedan.

In terms of safety, the ID. ERA 9X uses hot-formed steel with a strength of up to 1300MPa in key parts of the body; the battery pack has obtained IP68 dustproof and waterproof certification and has successfully passed rigorous tests such as extrusion, needle penetration, fire and water immersion.

Meanwhile, the vehicle comes standard with multiple airbags and side curtain airbags, and is equipped with a wealth of safety features such as vehicle stability system, tire pressure monitoring, hill start assist, and rollover prevention system, providing comprehensive protection for travel safety.

Data shows that in January 2026, Volkswagen's sales in China reached 89,600 units. Although its core market of traditional gasoline-powered vehicles remains solid, with cumulative sales exceeding 26 million units, the pressure it faces in the wave of new energy vehicle transformation continues to grow.

Previously, SAIC Volkswagen announced that it would launch seven new energy vehicles in 2026. As the flagship pioneer among them, the ID. ERA 9X not only shoulders the heavy responsibility of overcoming challenges, but also serves as an important verification platform for extending the cutting-edge technologies of the entire ID. family to more models.

This car represents a paradigm internally known at Volkswagen as "global technology + local needs." It's not simply a matter of transplanting German cars to the Chinese market, nor is it about blindly abandoning German engineering genes to cater to popular trends.

Behind the EA211 range extender lies years of deep accumulation in engine research and development; the 12-year anti-corrosion standard and 10-year aging test are engineering specifications that Volkswagen has long adhered to; and features such as the intelligent cockpit, rear entertainment screen, and four-zone voice interaction are localized choices tailored to the needs of Chinese families after a deep understanding of their car usage logic.

However, the market that the ID. ERA 9X is about to enter is already fiercely competitive. The 300,000 RMB-level large six-seater SUV segment is arguably one of the most competitive in the market today.

The Li Auto M9 has firmly captured consumers' minds thanks to Huawei's intelligent halo; the new Li Auto L9 vows to redefine the standard for home flagship phones; the NIO ES8/ES9 has built a solid moat by relying on its battery swapping system; and strong competitors such as the Li Auto LS9 and Voyah Taishan are also entering this red ocean market in their own ways.

So, why should the general public be able to get a share of the profits?

The answer might just be that slightly "old-fashioned" term in the context of new energy vehicle brands—German engineering. Among a host of domestic flagship phones that emphasize intelligence and technology, Volkswagen's brand DNA itself constitutes a differentiation: solid mechanics, reliable chassis, and rigorous engineering.

"German quality" remains a golden brand deeply trusted by Chinese consumers.

However, SAIC Volkswagen clearly understands that brand halo alone is far from enough. The ID. ERA 9X must demonstrate its true capabilities and fill in the gaps in its essential skills in terms of intelligence, space performance, range, and cabin experience before it is qualified to step into this fiercely competitive arena and compete with other emerging domestic brands.

#Welcome to follow iFanr's official WeChat account: iFanr (WeChat ID: ifanr), where more exciting content will be presented to you as soon as possible.

The story of AI has been told for two or three years now.

If the stories of the past two years have focused on upgrading large models, then starting this year, what we want to see is how models can step out of the computer screen and connect with our real world.

At the beginning of the year, robots performed skits, danced, and did kung fu at the Spring Festival Gala. This month after the New Year, OpenClaw has equipped all electric objects with AI's ability to learn and understand rationally.

According to the "Embodied Intelligence Development Report (2025)" by the China Academy of Information and Communications Technology, as of December 2025, there were 744 investment events in my country's embodyed intelligence and robotics field, with a total financing amount of 73.543 billion yuan, indicating a very broad application prospect.

For the average consumer, "robot" may still seem somewhat distant. But the robotic vacuum cleaners commonly found in many households are essentially intelligent cleaning robots.

Therefore, it's not surprising that I saw quite a few "robot" exhibits at this year's Appliance & Electronics World Expo (AWE).

However, when I saw this "puppy" at the Ecovacs booth, I stopped in my tracks.

This robotic dog, named "Fluffy," seems out of place among the robotic vacuum cleaners. It doesn't sweep or do housework. Its only job is to act as a cute little dog.

It was this little dog that made me realize that Ecovacs is telling a bigger story than just cleaning robots.

A puppy redefines "robot".

Companion robots are nothing new these days, but the fluffy robot's appearance is truly eye-catching.

With a Maltese-like appearance and covered in soft white fur, its two big, bright eyes look at you. In this exhibition surrounded by all kinds of steel and home appliances, the fluffy little thing looks even cuter, making people want to reach out and pet it.

When I stroked its soft head, it made a cheerful "woof woof" sound, wagging its head and tail in response to me with the same enthusiasm.

This is what makes Maotuan'er different. I'm hardly "manipulating" it; rather, I instinctively want to pet it, talk to it, and praise it, just like when I see my own dog. Maotuan'er will give corresponding feedback, tilting its head, wagging its tail, and blinking its eyes.

Such natural interaction is thanks to the multimodal perception capabilities behind the fluffy ball.

First, there are tactile sensors located on the furry ball's head, chin, ears, and back—these are "sensitive areas." Just stroking or patting the furry ball will make it very happy, and it can even recognize the amplitude and direction of the movement, giving different responses.

With a camera in its nose and microphones in its ears, the furry little dog has visual and auditory perception abilities, enabling it to recognize its owner's facial expressions, understand its commands, and provide corresponding emotions and feedback. For example, it will cheer and jump for joy when its owner praises it, and it will whimper and comfort its owner when the owner is sad and crying.

This design frees the furry friend from the mechanical "question and answer" interaction, and instead allows for a more realistic interaction style, closer to that of a real pet. A glance, a gesture, or a touch can all become part of the communication.

Besides the various feedback, I also love the furry friend's proactive interactions: when it sees its owner approaching, it will wag its tail happily to welcome them, and when it hears music, it will shake its head and sway to the beat.

To make the furry friend more authentic and further capture users' hearts, Ecovacs has also equipped it with a "personality system": it is born with five basic personality traits: gentle, sunny, clingy, sensitive, and irritable, as well as seven emotions such as happy, curious, and attached.

These settings may seem a bit abstract, but they are very intuitive in real-world interactions.

Through daily interactions, users can perceive the puppy's different emotional expressions from various feedback. As time goes by, the puppy gradually remembers the user's interaction habits and adjusts its behavior accordingly. Every action and word spoken by the user during their time with the puppy continuously shapes its personality.

Only by leveraging the capabilities of large models that can learn over a long period of time can Maotuaner truly shed its image as a mere "toy," transcending a machine that repeatedly performs pre-set actions, and becoming a partner that lives and plays with users as much as possible.

Emotional value is arguably the rarest "quality" in our time. How to make ourselves happy has become a major starting point for our consumption and decision-making.

A Gallup survey of more than 140 countries worldwide in 2023 showed that at least one billion people around the world feel lonely, with the highest proportion of loneliness among young people aged 19 to 29.

With the trend of atomization of families and individuals, "loneliness" has gradually become a fairly mainstream emotion, but "companionship" is not so easy to obtain; it requires more effort. Therefore, companion robots have emerged.

Through natural, predictable, and warm interactions, the fluffy ball provides instant emotional comfort. A glance, a response, a gentle touch—all these make people feel cared for and understood, something traditional electronic products cannot offer.

Compared to the humanoid robots I've seen in various situations before, which I "only dare to admire from afar and not touch," I'm more willing to hold this fluffy little thing in my arms, constantly stroking it, feeling the warmth emanating from its body temperature system, and enjoying its little movements and satisfied humming sounds.

For many more robots in the future, they will all need to learn from this little dog how to enter the sensitive inner world of us humans.

Ecovacs' comprehensive robot layout

Mao Tuan'er is just one aspect, an exploration from the emotional dimension. Ecovacs' home robot layout is a complete embodied intelligence strategy oriented towards function, learning, and management.

Essentially, Ecovacs' most well-known robotic vacuum cleaner products are also a kind of "integrated intelligence," but they are robots focused on the dimension of "cleaning." At AWE, we also saw products like the "Deebot X12," "Deebot T90 Pro," and "Window Cleaner W3."

But our vision of a "home robot" should not stop there; it should be a "butler-like" robot that can learn our needs, remember our habits, and manage our three-dimensional home space.

Ecovacs' answer to this need is the "Eight Realms" butler robot.

If a robot vacuum cleaner solves the task of cleaning the "floor", then "Bajie" is the intelligent butler of three-dimensional space, responsible for organizing, putting away, delivering, storing and assisting.

Simply "completing tasks" is not enough. Ecovacs believes that self-learning is an essential capability for a category of "butler robots." With the support of the recently popular "Lobster" OpenClaw intelligent agent, "Bajie" can not only passively execute instructions, but also learn and understand.

Traditional home robots, such as robotic vacuum cleaners, are just a collection of "functions" and still fall into the category of "tools." With the help of OpenClaw, Eight Realms does not exist to complete a single task. It has long-term goals and can make its own decisions based on the scenario.

A very specific scenario is that when the Eight Realms see an object on the ground, they do not simply grab it and place it properly. Instead, they try to understand and analyze what the object is, who it belongs to, where it is usually placed, and whether it is "suitable to grab" now.

Similar to Maotuan'er, Bajie also has the ability to "cultivate". The longer Bajie spends with users, the more it will transform "experience" into "memory", recognize each member of the family, and understand their different living habits and the placement of items.

Simply put, the longer Bajie interacts with users, the better it understands them and the more personalized it becomes.

When the Bajie, Maotuan'er, and Dibao are placed in a scene, Ecovacs' complete embodied intelligence picture is revealed: from tools to butlers to companions, different robots perform their respective duties and together form a system that serves family life.

The shift from single devices to multi-robot collaboration represents a significant direction for embodied intelligence in the home environment.

Intelligent robots in the home are no longer solitary devices, but rather multi-faceted collaborative teams – this is the future direction of embodied intelligence.

The robot walks into the home, Ecovacs is ready.

Elon Musk has declared that in the future, every household will have a robot.

Over the past two or three years, embodied intelligence has moved from laboratory exploration to the grand stage of galas, receiving the spotlight, and public awareness and acceptance have been continuously improving. However, a contradiction has gradually emerged:

Robots look cool, but what is their relationship with us?

In fact, for robots to enter our homes, the gap between "science fiction" and "reality" is far greater than the threshold between "technology" and "practicality."

Just like a robot vacuum cleaner, it is essentially an "intelligent cleaning robot" that targets the very specific scenario of house cleaning. It truly liberates humans from repetitive and heavy cleaning labor and successfully verifies the practical value of robots to humans.

For this company, whose full name is "Ecovacs Robotics", the robotic vacuum cleaner is just a starting point, just one part of its intelligent robotics strategy.

With the maturity of multimodal artificial intelligence technology, it is time for Ecovacs to move to the next stage. If in the past the core of "robot vacuum cleaners" was "sweeping," and "robot" was just an accessory; now, Ecovacs' goal is to move from a tool-type home appliance to a multi-dimensional intelligent partner.

Thus, at this year's AWE, we saw the Bajie as a "housekeeper" and the Maotuan'er as a "partner." They have gone further than the Deebot robot vacuum cleaner and are gradually enriching Ecovacs' "embodied intelligence" array.

From robotic vacuum cleaners to the Bajie and Maotuaner, each step represents a solid technological relay: the multimodal capabilities developed for obstacle avoidance, the machine learning algorithms created to learn the home environment, and other such technological accumulations are all natural and logical steps that Ecovacs takes towards embodied intelligence.

An embodied intelligence startup is bringing robots to the stage and introducing this new technology to the world, which is certainly a highly challenging technological breakthrough.

But to make the entire business model work, you need not only technology, but also insight. You need to take robots out of the spotlight and bring them into everyone's home.

When I saw many passersby in the AWE exhibition hall stopping to admire the furry creature and touching its head, I believed that the future of human-machine coexistence was already faintly visible.

#Welcome to follow iFanr's official WeChat account: iFanr (WeChat ID: ifanr), where more exciting content will be presented to you as soon as possible.

Yesterday, March 15th, the areas surrounding several problematic auto shows in Hangzhou, Wuhan, and Guangzhou nearly turned into scenes of physical altercations.

As soon as the car owner, who intended to enter the site to protest, drove his car to the entrance, he was surrounded by a group of unidentified people. Some people pleaded through the windows, some formed a human wall to block the way, and some even took advantage of the chaos to snatch the materials from the car owner's hands.

In a widely circulated short video, a young man wearing a mask stuck his foot under the front wheel in an attempt to stop a slowly moving electric vehicle.

Scamming someone in front of the camera is considered a staged accident.

To stop car owners who brought "problematic vehicles," some car manufacturers and dealers used almost every means at their disposal. Quality disputes that were usually delayed in after-sales service at the store were all moved to the roadside, turning into face-to-face pushing and pulling.

That evening, CCTV's 315 Gala aired on time. The two-hour program focused on discussions about bleached chicken feet, "miracle drugs," and AI-induced poisoning, while cars, the most expensive everyday purchase for ordinary people, were virtually invisible throughout the entire program. Those industry professionals who had tuned in beforehand likely breathed a sigh of relief.

However, landing safely on the biggest stage only makes some car companies more aggressive in handling customer complaints. It seems that as long as they hire people to block the car owners who are leading the protests from the venue and suppress the public opinion that day, the public relations task during this period is considered to be passable.

Unfortunately, public opinion can be suppressed, but reputation cannot be blocked.

The absence of cars at the 315 Gala does not mean the world is at peace.

The reality is that the number of customer complaints about car quality nationwide has reached a level that cannot be ignored.

According to the "2026 China Automotive Product Quality Trend Report" jointly released by CheZhi.com and Kairui Saichi, they received 227,803 valid complaints with real names in the past year, a year-on-year increase of 31.6%, breaking the historical record.

The staff sent by the manufacturer to the entrance of the problematic car show can only stop one car at most, while on the car quality website, there are hundreds of thousands of specific and minor quality disputes piled up.

Upon closer inspection, complaints about new energy vehicles are approaching half of the total. Within this category, plug-in hybrid electric vehicles, the most popular in the market, have become the hardest hit area for complaints.

At their product launches, major brands dedicate significant time to highlighting the advantages of plug-in hybrid technology, emphasizing that this solution eliminates range anxiety and boasts outstanding fuel economy. However, real-world driving experiences reveal a different story. The jerky transitions between the two powertrains and the precipitous battery drain in extreme weather have become a nightmare for many car owners.

The most unsettling data in the entire report is related to new cars.

Among car owners who filed quality complaints in the past year, 53.3% filed their complaints within 30 days of taking delivery of their vehicles. This figure seems to defy common sense; previously, people typically drove a traditional gasoline-powered car for several years before encountering minor problems.

I know times have changed, but haven't they gotten better?

The competition in the automotive market is becoming increasingly fierce, with all brands racing to launch new models. The development cycle for a new car has been compressed to a few dozen months or even less. Durability testing, which used to take years of meticulous refinement, is now often rushed and hastily completed.

Car manufacturing is increasingly resembling smartphone manufacturing. Many automakers choose to assemble the hardware first and deliver it, leaving software issues to be addressed through subsequent OTA updates. Even worse, some companies, in order to conceal battery safety hazards, secretly limit the vehicle's charging capacity and output power in the background, resulting in a reduction of the advertised 400km range by half after a single upgrade.

Car owners are paying real money and taking on unknown risks, only to end up helping the manufacturer find problems on their daily commutes. On top of that, they also have to be constantly wary of backstabbing from existing owners when a new model is released with lower prices and more features a few months later.

Most of the car owners who were turned away at the auto show were those who had bought these "half-finished products".

When they went to the store to communicate with after-sales service, the most common answer they heard was to wait for the next firmware upgrade. Unable to get their cars repaired and with their requests being passed around, people could only pin their hopes on problem car shows. These car shows usually didn't have fancy booths; the venues were very basic, just rows and rows of "problem cars" parked inside.

The car owners naively believed that driving their cars into the spotlight would bring them justice. They had no idea that waiting outside the exhibition hall weren't engineers with testing computers, but rather security personnel who had been pre-arranged. With the bottom line of product quality breached, even the last vestiges of dignity left between the buyers and sellers were quickly eroded.

A public relations battle trapped on the street

Looking back at past 315 Galas, cars have appeared on stage and even been the main focus.

In 2013, Volkswagen was singled out for a transmission problem, triggering an epic recall of over one million vehicles. At that time, the Chinese auto market was experiencing explosive growth, and bringing up this joint venture giant was meant to warn the entire industry against reckless expansion.

In the following years, the gala's cameras focused on the shady practices of after-sales service in dealerships. Several established car manufacturers were caught overcharging for repairs, and all issued apologies that very night, vowing to rectify their dealer networks.

Even in the past two years of transition between old and new models, the screen still shows popular cars that were discontinued due to quality issues, as well as luxury sedans that were publicly displayed because of abnormal noises from their driveshafts.

For a long time, the evening of March 15th has been the most agonizing few hours for all car company public relations teams. Everyone is glued to the television screen, terrified that their brand will become the target of nationwide condemnation.

Now the sword that was hanging over their heads has been put away.

The automotive industry has tacitly disappeared from the highest-level stage of consumer rights protection. With the withdrawal of central media oversight, the pressure has poured down on local governments like a flood, and numerous problematic car shows of all sizes have sprung up like mushrooms after rain. The problematic car show in Hangzhou has even reached its fifteenth edition, while those in Shenyang and Guangzhou have reached their fifth and sixth editions, respectively.

This is an extremely distorted industry ecosystem. Completely different from the auto shows you've visited before, these places lack glamorous booths, dazzling spotlights, and brand executives spouting sentimental speeches. The open space is simply filled with rows of cars, their windows plastered with various helpless slogans protesting consumer rights. Standing beside the cars are a group of ordinary people who have spent hundreds of thousands of yuan only to end up feeling utterly frustrated.

The scene at the Wuhan Auto Show was heartbreaking, with hundreds of car owners bringing problematic vehicles from over thirty brands together. Some new domestic cars had received numerous complaints due to chassis noises and brake failure in one-pedal mode, while some established joint venture vehicles were still plagued by the common problem of particulate filters clogging.

Faced with the spontaneous and concentrated exposure from the public, manufacturers' response has regressed to the most primitive stage. The reason they dare to block people from entering is purely because they have calculated their immediate gains and losses.

In today's market, regional managers often face very demanding sales targets. If a popular model is exposed at a problematic auto show, its sales next month could be greatly affected. Even if it costs more money to appease the leading car owners, it is much easier than admitting that the product is defective and launching a large-scale recall.

Moreover, the situation in this street battle is actually more complex than imagined.

Some car company employees even brought their laptops and drafted return agreements on the spot, lying on the hoods of the protesting vehicles. They blocked car owners from entering the premises and even made promises they could keep.

Give me ten minutes, and I'll take care of it for you on the spot.

Problems that you can't solve even after talking endlessly at a 4S store can be resolved with a full refund or even more compensation within ten minutes at a problem car show.

This practice of paying money to avoid trouble has distorted the issue of protecting one's rights. It sends a signal to the outside world that as long as you dare to make a big fuss and drive your car in front of the media, manufacturers are willing to pay money to suppress the heat.

In this way, what was originally a simple matter of protecting quality rights could potentially turn into a gray market profit chain. Everyone is speculating about the other party with the worst intentions and using the most extreme means to achieve their goals.

Buyers and sellers are caught in a cycle of mutual distrust, and after-sales service for automobiles, a capital-intensive commodity, has turned into a street farce where everyone is willing to go to extremes.

The annual gala ended on time, and the executives of the car company had a peaceful and uneventful night.

However, the problematic vehicles intercepted outside the exhibition hall, and the ever-growing number of complaints on consumer rights protection platforms, did not simply disappear. Using brute force to silence consumers might preserve sales for the next month or two. Automakers that rely on half-finished products and a power-driven logic will eventually pay the price under the very rules they set for themselves.

When ordinary users are forced to become rights defenders, this car manufacturing frenzy is nearing its end.

#Welcome to follow iFanr's official WeChat account: iFanr (WeChat ID: ifanr), where more exciting content will be presented to you as soon as possible.