The 2026 home companion robot track looks remarkably like the smart-hardware wave of 2014 — everyone is scrambling to seize the entry point, but the ones that truly survive will be the products people still use every day after the tide goes out. Mass production is only the starting point; retention is the endgame. In this cross-sector melee, whoever can make users still want to say a word to their robot three months later is the one closer to winning the race.
From the Corner to Center Stage
The 2026 World Artificial Intelligence Conference (WAIC) sent a clear signal: the home companion robot is no longer a niche category tucked away in a corner. More than 200 embodied-AI companies and over 3,000 exhibits — behind these numbers, "robots entering the home" has moved from concept to a shared narrative across the entire industry.
But unlike the smart-hardware wave of 2014, this round's entrants come from more mixed backgrounds, and their routes diverge far more sharply. An in-depth report by Guanchazhe Wang (Guancha.cn, a Chinese current-affairs news portal) maps the entire track: humanoid robots, plush desktop pets, industrial giants pivoting to consumer products, emotional-software companies crossing over into hardware — everyone wants to occupy the living room, yet on where to start, how to make money, and why users would pay, there is no consensus answer to any of the questions.
Two Technical Routes, One Shared Anxiety
A clear dividing line runs through the track.
The humanoid robot route is represented by Unitree Robotics — from the R1 to the G1 to the H2, covering the consumer market from lightweight to full-size machines. UBTech Robotics has launched the sub-brand "Youshijie," whose U1 series spans from a half-body version at 119,800 yuan to a flagship at 990,000 yuan — a price spread of nearly tenfold. The collective influx of automakers makes this route more crowded still: XPeng, Tesla, Xiaomi, BYD and nearly 20 other carmakers have already entered through in-house development or investment. The problem is that humanoid motion control and general-purpose manipulation remain at an early stage in home environments, while the high price tag keeps most consumers at bay. UBTech founder Zhou Jian announced 13,361 pre-orders — but a closer look at the rules reveals that most of those orders were placed before prices were announced and can be refunded unconditionally before the balance is paid. The market will judge for itself how much gold is in that number.
The non-humanoid companion route has chosen a different point of entry. Ludens AI, founded by former DJI engineers, offers the desktop companion robot INU and a mobile version, COCOMO, emphasizing non-verbal emotional interaction and on-device AI — running fully offline and resolving privacy at the architecture level. Ecovacs, the robot-vacuum leader, has launched FAMIBOT LilMilo ("Little Furball") at $599: plush biomimetic material, a constant surface temperature of 38 degrees Celsius, a built-in large language model with a long-term memory module, more than 40 sensors, recognition of 21 fine-grained emotional states, and 360 minutes of battery life — reusing the sensor technology, mass-production capability, and understanding of home spaces accumulated from robot vacuums. Fourier Intelligence's GR-3 targets companionship in eldercare and rehabilitation. Newcomer Shouxing Technology has rolled out its ultra-biomimetic Elf series.
The two routes each have their partisans, but they share the same anxiety: the enormous gap between task-completion rates in the laboratory and performance in real home environments. This is the core contradiction highlighted in Stanford's AI Index Report 2026, and it is a consensus held by virtually every practitioner — it is just that few are willing to admit it on stage at a launch event.
The Cross-Industry Entrant's Unique Asset: Thirteen Years of Emotional Data
Among all the entrants, Xinyan Group is the one that looks least like a robotics company.
It did not start in hardware, nor in robot bodies, nor even in AI large models — it grew out of an emotional-companion app. Its Cece app has been online since 2013, serving a cumulative 66 million registered users with more than 27,000 contracted counselors, and has amassed an emotional-conversation corpus on the scale of trillions of tokens. Its in-house Xinyuan large model passed the filing with the Cyberspace Administration of China in May 2024 — currently the only vertically specialized large model with a national-level filing in the field of emotional counseling and companionship.
At this year's WAIC, Xinyan gave the first offline showing of its embodied-AI product Bubbo — about 65 centimeters tall, plush material, anthropomorphic design, with mass production and shipments planned for the end of the third quarter of 2026.
What is distinctive about this path is this: the core moat of a companion robot has never been its mechanical structure, but the depth of its understanding of users' emotions and the naturalness of its interaction. Hardware can be sourced externally, joints can be customized, skin material can be found from suppliers — but "understanding why a person suddenly feels sad at three in the morning" has no shortcut; it can only be fed by data. Thirteen years of emotional-conversation data from 66 million users is an asset that pure hardware companies cannot replicate in the short term.
Yet the problems are just as obvious: when a software company builds hardware, supply-chain management, quality-control systems, and after-sales networks are essentially brand-new subjects. Whether users of the Cece app are willing to pay for a physical robot remains to be tested by the market. Xinyan Group's approach is to avoid the humanoids' weak spots in motion control and general-purpose manipulation and concentrate its limited compute on emotional understanding and proactive interaction — the cleverness of this strategy lies in choosing the battlefield where rivals are weakest and it itself is strongest.
The Core Contradiction: A Fracture in the Purchase Decision
A follow-up question quoted in the report is the core contradiction of the entire piece: being willing to spend money on emotional value, and being willing to spend thousands or even tens of thousands of yuan to bring a robot home, are entirely different decisions.
The structural fracture facing the industry today is this: product capability is still at an early stage, yet pricing and narrative are already telling a mature-market story. Ecovacs pricing LilMilo at $599 is an interesting data point — entering the companion-robot field at a robot-vacuum price band, trying to lower consumers' barrier to a first purchase while reusing the robot-vacuum channels and supply chain. But the renewal logic of a companion robot is entirely different from that of a functional appliance: the value of a robot vacuum lies in "mopping your floor for you every day," while the value of a companion robot lies in "you being willing to talk to it every day." The former is certain; the latter is full of variables.
A Plain Question
The report closes with a question whose quality surpasses that of most industry analysis.
If a consumer with real willingness to pay buys a companion robot and puts it in the living room, will they still turn it on three months later?
The question is precise precisely because it sidesteps every contest over technical parameters — number of joints, realism of the skin, scale of the AI model — and goes straight to the endgame competitive factor of the companion-robot industry: retention. Whether users keep interacting with it after the purchase. The answer to this question determines whether a companion robot is a "one-time novelty toy" or a "long-term family member."
On this point, the depth of thirteen years of emotional data may be more persuasive than 88 joints.
Guancha.cn in-depth report on the home companion robot track at WAIC — 2026-07-21 11:10