王与桐

高阳 defines embodied intelligence as an industry that has moved from open-ended research into engineering convergence, rather than a distant concept still waiting for a flash of inspiration. In 2023, he felt the room for truly groundbreaking research was narrowing, so he chose entrepreneurship as a way to engineer mature paradigms into products. To an outside world that still sees a jumble of competing approaches, his view is: “The ingredients are all there; it just takes time to build it.”

The industry’s key capability inflection point is L2, not bipedal locomotion or flashy single-task demonstrations. 高阳 defines L2 as a robot that can move around a single setting such as an office and complete roughly 5-8 tasks; L3 means covering 70%—80% of the tasks humans can perform in that setting. At the time, he expected L2 by “next June,” with L3 requiring another 1.5-2 years; the serviceable market could expand 10x or even several dozen times from L1 to L2.

The core technical path is end-to-end VLA paired with fast and slow systems, while the priority form factor is a wheeled base with 2 arms. The slow system understands vision, language, and intent; the fast system processes tactile and low-level visual feedback at 50—100 Hz. Bipedal robots will eventually need to be added, but 高阳 argues that “the main bottleneck is what your hands can do”; wheeled dual-arm robots may already cover roughly 80% of the scenarios robots can solve.

Data scale will be the key constraint on embodied-model scaling. 高阳’s team has observed a log-linear relationship between data volume and the optimality gap. Roughly speaking, “collect 10x the data, and performance gains another 9.” Following its roadmap to an embodied “3.5” would require roughly 10B useful internet videos, 100M teleoperation trajectories, and tens of millions of RL trajectories, over an estimated 4-5 years.

Folding laundry has become the industry’s shared demo because it tests deformable-object state understanding and generalization at the same time, not because housework is the first commercial use case. A shirt’s wrinkles cannot be reduced to a handful of fixed states: the robot must first understand “how it got messed up,” then decide how to unfold and fold it. The next hurdle is to change the fold on command or hide an envelope under the left sleeve—multi-task combinations that mark the transition from a single-task “IQ test” to L2.

China’s immediate edge is hardware manufacturing, repair, and the speed of experimental iteration, but the biggest value still lies in integrating the AI brain with the body. Robots in Tsinghua’s high-intensity experiments may be shipped back to Hangzhou every week, while 千寻’s in-house engineers typically resolve most issues within a day or 2. Even a US team that bought 100 robots could find that “the pace of repairs is hard to keep up with the pace of experiments”; but building only the brain leaves the system without the body-specific “muscle memory” it needs.

Commercialization should be delivered progressively with capability levels, rather than demanding that a prototype robot immediately function as general-purpose labor. 高阳’s response is that “you cannot demand GPT-1 have commercial capability while it is still GPT-1”: today’s L1 can enter a limited number of high-volume industrial stations such as automotive final assembly, and some customers are willing to pay RMB100,000-plus for a robot. The company’s real priority should be to reach “GPT-4.0”-level capability while using limited commercialization to improve resilience.

Li Yong’s core view is that children will be the first segment where AI hardware finds real product-market fit, with plush toys the best vehicle for now. After Tmall Genie sold 30 million units, the team unexpectedly found that “children interacted more”: adults demand too much from AI tools, while 3- to 6-year-olds naturally talk to characters. Plush toys already carry emotional value and a defined persona—“it’s just a 4-year-old Peppa Pig; what happens if it gets an answer wrong?”—allowing the product to exploit LLMs’ natural-language capabilities while keeping first-wave user expectations in check.

Hiweaver’s shift from educational hardware to AI plush was not trend-chasing, but a deliberate move away from a battlefield where education giants would inevitably prevail. The company once shipped hundreds of thousands of early-learning devices and dictionary pens, only to face homogenization and price wars after Yuanfudao, TAL Education and others entered hardware. Even replacing Tmall Genie with ChatGPT would leave models, courses, curriculum development and brand-building areas where the giants would eventually catch up. The team therefore chose the “emotional-value track,” reducing serious education to light functions such as hydration reminders, courage-building and interest discovery.

The first BubblePal validated demand through existing plush-toy pendants and has sold tens of thousands of units, but explosive PMF also exposed delivery weaknesses caused by limited capital. The product officially launched in August 2024, with Douyin livestream peaks above 3,000 viewers and thousands of customer-service inquiries a day, while the company had only 1 livestream host and its app could not even reach domestic app stores in time. Before launch, the company relied on bank loans to stay afloat. Financing improved materially after sales data emerged, but Li Yong still judged the overall operation that year as “below expectations.”

Hardware products are won first through trade-offs: BubblePal sacrificed wake-word activation and some convenience for lower cost, faster launch and character-based conversation. The 720mAh pendant needs charging roughly once a week and requires users to hold a button while speaking; the full plush version is planned to use a 3,000–4,000mAh battery, support far-field interaction and last about a month per charge. After tens of thousands were sold, the 2 biggest complaints were the need for an internet connection and app download, followed by “having to hold the button,” but Li Yong insists the first generation should prioritize role-play, continuous conversation and long-term memory.

Children are only the entry point: Hiweaver is betting on young people repeatedly buying multiple AI characters, rather than on one-off children’s hardware constrained by birth rates. Li Yong says the children’s line may account for less than 20% of leading plush brands’ business; the main consumers are young adults. Children came first because 2023–2024 models were not yet good enough to support adult purchases of emotional-value hardware. The inflection point is the experience and cost of end-to-end voice. Once mature, the company hopes to build an “AI version of Jellycat” or a “Toy Story world,” while its 2023 move to Shenzhen was partly aimed at entering the US market and addressing geographic limits.

The moat is not access to a particular foundation model, but the long chain spanning software, hardware, algorithms, IP, supply chain, design, marketing and user insight. Li Yong estimates that “at least 100 teams” could hand-build a batch of AI toys, but getting the experience to 80 out of 100, selling the products and building a lasting brand are entirely different tasks. China’s supply chain can manufacture Jellycat-style plush, but it does not automatically create Jellycat: marketing, taste, design and insight into user scenarios also determine the premium. Tool categories are like chess and often tend toward winner-take-all; toys are like Go, where multiple IPs and brands can coexist.

The next upside leg will come from GPT-4o-style end-to-end voice, multi-agent systems, multimodality and on-device models, while the main risks have shifted from technology to organizational execution. Hiweaver has integrated DeepSeek V3 but has not adopted R1’s chain-of-thought because voice use cases prioritize latency. Planned 4G, cameras, multiple characters interrupting one another and environmental sound are all constrained by cost and chip compatibility. After financing, the key questions are overseas expansion, user operations, parallel R&D and hiring discipline—Li Yong worries that the linear assumption that “20 people can build 2 products, so 40 people can build 4” will be undermined by bureaucracy, attrition and diluted values.