做了十年情感服务，他为什么非要造一台机器人？|甲子光年

After Ten Years in Emotional Services, Why Does He Insist on Building a Robot? | JiZi Light‑Year

Embodied intelligence entering homes may truly start with emotional companionship.

Author｜刘杨楠

Edit｜Wang Bo

“Dad, what is a black hole?”

Faced with his daughter’s curiosity, Ren Yongliang didn’t brush it off. He explained the principle of black‑hole collapse carefully—scientific, accurate, and airtight. Yet in the following nights, he noticed his daughter was afraid to sleep—she feared being sucked into a black hole.

This incident gave Ren a stark wake‑up call. He realized that communication isn’t just answering every question; some unspoken empathy and sense of proportion determine whether a conversation truly reaches the other person. After more than a decade of providing emotional services, he still felt at a loss in everyday talks with his child.

Ren Yongliang is the founder of XinYan Group. In 2001, Ren entered Peking University Medical School; in 2011 he left IBM to start his own business, and in 2013 launched the CeCe app. Currently, the CeCe app has accumulated over 66 million registered users, serves nearly 2 million users daily, and has provided more than 200 million minutes of real‑person emotional companionship over ten years.

Now, he has a new role—founder of XinYan Robot and chief designer of Bubbo 1.

On October 29, 2026, XinYan Robot officially released “Magical Companion” Bubbo 1, a product nearly two years in the making. Bubbo stands 65 cm tall, offers a perspective similar to a two‑ to three‑year‑old child, features a plush shell and round eyes, and hides 13 types of sensors for a total of 26 sensors. Bubbo is now on sale on JD.com for ¥9,499; the first 1,000 users can enjoy a discounted price of ¥7,999.

When the tide of embodied intelligence arrives, Ren Yongliang is pushed toward the next stage of emotional companionship. He tries to give a more powerful AI a body, using an outsider’s perspective to engage more deeply in family activities, allowing robots to fill the inevitable helplessness and regret in current family relationships.

Founders usually largely determine a company’s character. In Ren Yongliang’s narrative, “Jiazi Light‑Year” reveals a rare pragmatism in the embodied field today: he does not repeat the grand narratives dominant in the sector, he knows the strengths of XinYan’s capabilities, and he retains a reverence for the new wave of technology.

Embodied intelligence entering the home may truly begin with emotional companionship.

Image source: Xin Yan Robot

1. Looking for a hammer while holding a nail

Ren Yongliang’s entrepreneurial journey has been constantly corrected by reality.

In 2013, the CeCe app launched, positioned as an AI recommendation tool, but the technology at the time didn’t allow it. After mobile payments became widespread in 2016, Ren Yongliang turned it into a two-sided emotional counseling platform. Most users are women who seek advice on various relationship issues, and the platform connects them with experts and counselors to answer their questions.

Ren Yongliang realized that providing emotional companionship with real people is one of the hardest businesses in the world. Education still has a single standardized test, but life’s answer sheets are all different, making demand hard to standardize. The supply side is equally scarce—a great counselor needs talent and more than ten years of professional training, and can only serve one person at a time. “So I never advise anyone to start a business in this field,” Ren admitted.

The turning point came in 2016.

The year AlphaGo defeated the world Go champion, Ren decided to use AI to solve problems in the emotional companionship field. A large part of psychological counseling is language technology, and AI gave this toughest business a breakthrough point. Previously, the platform had amassed a wealth of real conversation records from numerous experts and counselors providing online services, and with professional annotations from psychology specialists, these could theoretically serve as high‑quality reference data for training AI.

These are the “nails” that Xin Yan has been accumulating for over a decade. Yet for nearly ten years, these nails have never found a “hammer” that could put them to full use.

In November 2017, Ren Yongliang announced an all‑in on AI at a company-wide meeting. To prove it, he even showed a slightly rough PPT from that year.

In his ten‑year technology roadmap, “home physical robots” were placed at the farthest end. At that time, the industry had not yet popularized the term “embodied intelligence,” let alone “dexterous hands.” Each project on the roadmap was labeled with a difficulty number, and Ren considered the “robotic hand” (dexterous hand) the hardest. History has proven his judgment correct—today even Elon Musk has not fully solved the dexterous hand problem.

Image source: Xin Yan Robot

But before the company could even deploy robots, Ren Yongliang already hit setbacks at the modeling stage.

In 2019, XinYan developed a 110‑million‑parameter QA model, but limited by the technology of the time, it could only handle a single round of questions. With the proliferation of large‑language‑model technology in 2023, Ren Yongliang was very excited; the internal infrastructure was already in place, and smarter models were integrated, possibly ushering in a new wave of user growth. Yet at that time, most companies had not yet found a true breakout point, and the growth of app penetration lagged far behind the improvements in model capabilities.

This is the dilemma faced by many AI‑to‑consumer application companies. The root cause lies in the huge perception gap about the AI wave between the industry and the outside world; many users who are not focused on AI or are far from the internet sector find it hard to quickly experience AI’s power.

Rigidly stuffing a large model into a phone is like running a high‑speed train on a paved road. To break this situation, we need to find ways to further lower interaction barriers and expand the potential user base.

Technology sets the lower bound of experience, interaction sets the upper bound—this is Ren Yongliang’s latest insight, and also the underlying principle of his design for Babu.

Image source: “Jiazi Light‑Year” photography

Every wave of technological democratization in the information industry has been accompanied by a shift in interaction paradigms. The personal computer was invented as early as 1975, but it only truly entered millions of homes after the graphical interface appeared in 1984; smartphones capable of internet access were born in 1996, and it was the iPhone in 2007, with its multi‑touch technology, that truly ignited the mobile internet.

So, after the birth of ChatGPT in 2022, what interaction mode can make AI capabilities widely accessible?

Some bet on brain‑computer interfaces, some develop portable wearables, others create desktop gadgets. The answer offered by XinYan is “partnered interaction” — let the product seek the user, rather than the user seeking the product. Achieving this shift requires a brand‑new form beyond phones and PCs.

In 2025, XinYan Robotics was founded, focusing on the development of emotionally intelligent home robots. In February 2026, Babu made its debut at the Beijing Radio and Television Station’s Spring Festival Gala; in mid‑August, the first mass‑produced Babu rolled off the line. After more than a decade of deep work in emotional companionship, XinYan has now reached the point where its value can be realized.

To many, emotional companion robots for the home are becoming crowded, and short‑term results seem unlikely. Many peers have warned Ren Yongliang to be cautious, not to jump into hardware lightly, and certainly not to rush into embodied robots for the home scenario.

But more than a decade of observation and reflection leads Ren Yongliang to conclude that the home is one of the most urgent needs and the scenario most aligned with current trends.

On the demand side, care and companionship for “the elderly and the young” have become a hot social topic. The Ministry of Civil Affairs’ 2025 report shows that the national population aged 60 and above has reached 323 million, accounting for 23% of the total population; nearly 60% of seniors are empty‑nesters, and more than 38 million live completely alone. The home‑care pain points of this group are highly similar—forgetting to take medication, falls that go unnoticed, and being unable to say even three sentences a day. On the other side of “the elderly and the young,” dual‑income families face intense pressure on childcare time, and emotional gaps persist between parents and children living apart, a long‑standing issue.

On the supply side, large models continue to iterate, and humanoid hardware costs have dropped below ten thousand yuan, providing an objective basis for a companion robot to enter households.

Policy is also driving this forward. In March this year, the Government Work Report listed “optimizing services for the elderly and the young” among the 25 livelihood projects for the start of the 15th Five‑Year Plan; the State Council’s approved “Expanding Consumption 15th Five‑Year Plan” further called for strengthening technological empowerment and promoting digital intelligence upgrades in the elderly‑care sector. Embodied intelligence has been included in the Government Work Report for two consecutive years.

Therefore, for Ren Yongliang and Xin Yan, the best way to tap into the embodied wave may be emotional companionship in the home scenario. Babu was not hastily created by Xin Yan under FOMO or a chase‑the‑trend mindset; it is the realization of value after more than a decade of deep work in the emotional companionship field.

2. Partner‑style Interaction,

Redefining the Boundaries of Proactive Intelligence

Partner interaction is the soul of Babu, and Ren Yongliang defined four characteristics for it.

First, Babu must have an independent entity.The phone is only turned on at certain moments, and once it is, countless products compete for the user’s attention. However, a “partner” does not seize the user’s focus; it follows the flow of attention and provides appropriate services, thus requiring an independent carrier.

Second, Babu must have autonomous mobility.This is a highly controversial requirement; many people advise Ren Yongliang not to try it lightly, as the home environment is too complex. But his judgment is that if Babu cannot walk or switch perspectives, it cannot “see” more information, nor can it build sufficiently accurate and timely context, and its interactive capability would have nowhere to manifest.

With the first two items, Babu can achievethe third feature—“real‑time presence.”Its vision, hearing, and various perceptions must capture environmental changes at every moment, and also monitor changes in all people within the environment.

Finally, and alsothe most critical part, is turning perceived environmental changes into “semantic events”—converting continuous physical signals into clearly describable occurrences, such as who opened the door, who came back, what the person’s current emotion is, and so on. In Ren Yongliang’s view, “as long as you turn physical-world happenings into information that a large model can understand, you can create great products.” This fundamentally differs from voice‑command interaction; its context no longer depends on the user actively providing it, but is autonomously collected by the robot.

Ren Yongliang uses the currently popular personal agent as an analogy; he believes Babu is essentially a family agent (an embodied household intelligence). The former emphasizes the output side: the user gives a command, and it efficiently completes the task and handles complex matters;the family agent emphasizes the input side: the user need not issue commands, it can autonomously observe and perceive real‑time changes in the user and the surrounding environment, proactively telling the user what is needed at the moment.

The intrinsic difference between the two lies in the shift of the interaction perspective. Ren Yongliang jokes about himself: “When I’m in a meeting at the company, I think I’m already very gentle and pleasant, but employees might feel I’m too strict. Often, an external viewpoint can help us understand ourselves better.” The Family agent injects an independent external perspective into the emotional companionship of the home scenario.

To turn partner‑style interaction into a product, Ren Yongliang “forced” himself to build a piece of hardware from scratch. Because existing robot sensors are too few to support the powerful perception system required for partner‑style interaction. As Steve Jobs once said, to create the software you want, you have to make the hardware yourself.

Thus, Babu became the “companion robot with the most sensors,” equipped with 13 categories and 26 sensors in total. These sensors are not just a simple pile‑up; they together form the physical foundation of Babu’s perception, including vision, hearing, touch, and even behavioral trajectories, continuously collecting physical signals from the home.

Image source: photographed by “Jiazi Light‑Year”

And the one that turns these signals into intelligence is the Bubbo OS, which was fully unveiled for the first time at this press conference.

Bubbo OS differs from narrow‑scope operating systems like Linux or Windows; it is a perception‑focused Harness architecture whose core task is to convert sensor data from the physical world into semantic information that large models can understand.

To aid comprehension, XinYan borrowed Figure’s Helix hierarchical naming: S1 and S2 are taken from psychology’s fast and slow thinking, S0 handles real‑time control, and today’s industry‑hot topics such as VLA and world models generally fall within the S2‑to‑S0 range.

Built on this framework, Bubbo OS adds a self‑developed S3—perception layer—that converts multi‑sensor, multimodal real‑time physical signals into discrete, meaningful semantic events. Among the four stages—perception, memory, reasoning, and execution—perception has the highest priority.

Image source: Xin Yan Robot

Ren Yongliang explained that whether the context is sufficient determines the ceiling of current AI conversational ability, and enriching the robot’s perception greatly increases the amount of contextual information, thereby enhancing the model’s conversational capability. Therefore, the first-generation Babu product entrusts users with confirming and labeling multimodal information such as facial images, voiceprints, and behavior trajectories, which is a prerequisite for the robot to proactively and intelligently learn a sense of proportion.

Supporting this system is the XinYuan large model. Emotion data annotated by psychological experts is directly reused for conversational ability, and in the long term it will become a unified model with both psychological capabilities and real-time interaction capabilities. XinYan also collaborates with Harbin Institute of Technology’s Emotion Computing Lab, conducting in-depth research on personalized dialogue technology and a feedback loop for emotional data.

Based on this hardware‑software architecture, XinYan, focusing on the emotional companionship needs of two user groups—“one elder and one child” in family scenarios, has designed multiple functions for Babu.

For the elderly, it is first a caregiver.

When Babu detects an elderly person has fallen, the app willsend an emergency notification; children can view it remotely and even control Babuto approach the elder and confirm the situation; the heart‑rate monitoring feature will also link with the CèCè smart ring, so anyone at home who didn’t sleep well or has emotional fluctuations,Babu can sense it in time.

Beyond caregiving, it is also a listener. Ren Yongliang grew up with his grandfather; as a child he was tired of hearing stories about the Liberation War, and when he got older and tried to retrieve the details, he found many memory fragments had gradually faded. Ren believes that every family’s history deserves a rescue excavation. Babu will accompany the elder in revisiting the past, piecing together the memory fragments in the elder’s mind into a coherent whole.

For children, Babu’s primary mission is to catch their boundless curiosity and emotional needs.

For those “simple yet hard-to-answer” questions from kids, Babu can provide answers using a large model fine‑tuned by a professional team. The first generation of Babu comes with bilingual capability by default and features a unique tiered dialogue mechanism that delivers responses at different levels according to each family member’s comprehension; for more demanding users, it can even operate at the “person” granularity, enabling a “cultivation mode” for specific members that replies in a foreign language to any question.

Beyond “the elderly and the child,” Babu also looks after other family members. Its body height is set at 65 cm, roughly the height of a two‑ to three‑year‑old child, allowing Babu to intuitively capture the world through a child’s eyes and take family photos with no one left out.

In addition to facilitating timely communication with users, the accompanying app can also serve as “Babu’s Moments,” allowing users to see what Babu captures at home whenever they are out. It also compiles what it hears and sees into a report for the family, covering the child’s learning achievements, new interests, the partner’s mood fluctuations, and stories told by the elderly. Each report is a skill; in the future, XinYan will open up this capability so everyone can design their own reports, truly achieving personalized emotional companionship.

Beyond the core functions, XinYan also sets strict safety red lines for Babu.

In terms of materials, all fabrics and hats meet the GB 31701 Class A national standard, making them safe for children under three; the battery uses lithium iron phosphate and is CQC certified, with the cells passing 3C certification. “A robot entering a home leaves no room for physical safety compromises,” Ren Yongliang emphasized.

The privacy solution is a combination of edge and cloud. Except for content that users actively view, video information is first processed irreversibly on the device before being uploaded to the cloud. Intelligent imaging also must first be judged safe before it is sent to the app, and all image content can be deleted by the user at any time. For content security, multiple anti‑addiction options built to national standards are included, and we are cooperating with Harbin Institute of Technology on model security capabilities in the field of affective computing.

Beyond the architecture and features, this robot also has several counter‑intuitive details that best reflect Ren Yongliang’s product philosophy.

To prevent seniors and children from becoming overly dependent on AI products and affecting real‑world family relationships, Ren Yongliang deliberately halted the team’s optimization of response speed, giving Babu’s replies a certainsense of delay and breathing, avoiding an excessive pursuit of rapid and exhaustive responses, and aligning better with the natural rhythm of human‑to‑human communication.

At the same time, the XinYan team fully respects Babu’s “agency.” It does not need to unconditionally cater to the user; if a conversation goes on too long or the battery is low, it will automatically navigate to charging, not only saving the user repeated charging steps but also maintaining a proper sense of timing in the interaction.

Objectively speaking, the emotional companion robot field is still in its early stages, and there is no recognized successful case for consumer‑facing companion robots to date. Therefore, it is difficult for outsiders to accurately assess the strengths and weaknesses of various technologies or products, but Ren Yongliang’s design philosophy determines Babu’s uniqueness—whether for Xin Yan itself or for the entire emotional companion track, Babu is an irreplaceable presence.

3. Understanding the Human Heart Is the Key to Good Companionship

Currently, most family companion products are just plush toys wrapped in a large model shell—lots of concept talk, little actual interaction. This field has never lacked players; it lacks people who are serious about solving the problem.

What sets XinYan apart is a product philosophy that starts from “understanding people.” They aim to grasp the inner world of users and treat interaction itself as the main battlefield.

“Understanding the human heart” can be explored at different levels. Professor Zhao Yanyan from the School of Computer Science at Harbin Institute of Technology breaks emotional computing into three layers:

Emotion perception: detecting whether a person is happy or not; emotion cognition: understanding what’s behind it, what events or reasons; emotion interaction: providing substantive responses and assistance. Each layer is harder than the previous, and interaction with physical entities is the highest level. She also points out that multimodality, mobility, and proactive perception are unavoidable paths, which also confirms from an academic perspective that XinYan’s “high perception” approach did not arise out of thin air.

XinYan has spent over a decade getting close to users’ inner worlds. The TestApp has accumulated more than 66 million registered users, serving nearly 2 million users daily, and in 10 years the real‑person emotional companionship service has exceeded 200 million minutes. Yet moving from understanding the human heart to truly intimate companionship still faces many challenges, spanning the entire chain of robots entering homes.

When a robot enters an unfamiliar household, it must first establish emotional links with each family member, getting to know everyone from scratch, building trust, and gauging boundaries. There is no clear answer to how long this process takes.

After establishing an initial emotional connection, the boundary of proactive intelligence also remains unresolved. Babu can perceive, judge, and initiate interactions, but exactly what it should manage and what it should not is something even Ren Yongliang admits that Babu still needs continuous iteration.

In this process, whether Babu’s iterations can form an effective data loop with real user feedback remains a very tricky issue. Xin Yan has accumulated over a decade of online textual data from human‑to‑human interactions, while the home scenario generates multimodal data such as facial expressions, gestures, and spatial distance. How to bridge online and offline and keep evolving will require more time to explore.

Privacy and data security issues run throughout. The questions a robot must answer to truly become part of a family’s daily life are currently addressed by Xin Yan with an edge‑cloud integration solution, which still needs ongoing validation.

These challenges repeatedly remind all players that family emotional companion robots cannot remain merely at the level of capability demos and conceptual ideas; they must pragmatically confront the friction, boundaries, data, and security issues inherent in real family relationships, turning “can accompany” into “reliable everyday companionship” through continuous iteration.

From the pricing of this Babu model, XinYan Robot aims to exchange a very low price for valuable real user feedback, and then continuously iterate with one to two upgrades per month, making emotional companionship a constantly evolving service.

At present, XinYan is one of the few companies in the industry that truly understands the field and is willing to patiently refine technology and products with a pragmatic attitude. Its entry sends a clear signal to the crowded emotional companion arena: to excel at emotional companionship, you must first understand the human inner world; this tests not only technical and product expertise but also patience for long‑term investment.

(Cover image source: “Jiazi Light Year” shooting)

END.