On one side, a scientist-founder emerging from the research frontier; on another, a logistics veteran who's been through large-scale industrial deployment; and a technologist who's traveled the road from autonomous driving to Embodied Artificial Intelligence. Three founders from different backgrounds, standing on the front line of embodied AI, brought answers that don't entirely overlap.
How should world models actually be built? What should scaling really be measured against? How should technical trade-offs be made? And what kind of people does embodied AI entrepreneurship need most today?
This episode comes from a panel discussion at "The New Wave of Embodied AI: World Models and Intelligent Infrastructure," a special event co-hosted by Yunqi Capital and SEE Fund during the 2026 World Robot Conference. Three new partners Yunqi welcomed in 2026 — Huazhe Xu, founder of Poke Robotics; Sam Li, founder and CEO of Quantum Dynamics; and Nianqiu Liu, founder of PHYMI — joined Yu Sang, Executive Director at Yunqi, to share their most direct judgments and firsthand impressions of embodied AI, spanning model training, data, systems, products, and real-world commercialization.
· Opening remarks
Linda (Na Li), Managing Director, Yunqi Capital
· The panelists
Yu Sang, Executive Director, Yunqi Capital (panel moderator) Longtime follower of frontier technologies including AI and embodied AI
Huazhe Xu, Founder, Poke Robotics A scientist-founder with a long-standing focus on robot learning, world models, and general-purpose manipulation, now leading Poke Robotics in exploring general-purpose robots for home scenarios
Sam Li, Founder & CEO, Quantum Dynamics A veteran of industry and logistics automation, formerly CTO of Cainiao Logistics, with deep experience in unmanned warehouses, robotics, and large-scale automation systems. He now starts from logistics scenarios to drive the scaled commercialization of embodied AI
Nianqiu Liu, Founder & CEO, PHYMI Former technical partner and board member at DeepRoute.ai, where he experienced firsthand the critical phase in which China's advanced autonomous driving moved from L4 R&D and operations to scaled mass production. He now focuses on open-world scenarios, mobility, and Physical Intelligence
· Episode timeline
02:02 From demos to doing real work at scale — where are robots still stuck? Generalization of foundation models, end effectors, system-level closed loops, and the mobility needed to enter open-world scenarios.
08:02 What exactly is a world model? What's being over-hyped? From World Model to World Action Model: if the ultimate purpose of predicting the world is to change it, shouldn't that be the goal?
13:00 Does embodied AI have its own Scaling Law? Scaling isn't just about stacking parameters and data hours. Data diversity, quality tiers, and multimodal information may matter more than absolute volume.
19:39 How do frontline founders choose their technical paths? From data closed loops in living spaces, to real commercialization in logistics, to running domestic chips alongside NVIDIA chips — every company is making its own trade-offs.
23:02 "Let the customer speak": what does real deployment look like? Proving yourself not through benchmarks and flashy demos, but through cycle time, success rates, human takeover rates, and the economic value per task.
25:18 What kind of people does embodied AI lack most? "Expensive is fine, but pretentious isn't," rapid learning and curiosity, full-stack capability, and the ability to reason from first principles.
30:28 Will robots have their "ChatGPT moment" three years from now? How far will homes, logistics, and open-world scenarios each get? When will robots move from "novelty purchases" to "paying for real utility"?
🏢 About Yunqi
Founded in 2014, Yunqi Capital is an early-stage lead investor focused on digital intelligence and hard tech. We have been named a "China Top 10 Early-Stage Investment Firm" for multiple consecutive years by Zero2IPO, ChinaVenture, 36Kr, and other institutions.
Over the past 12 years, we've grown alongside more than 200 tech startups, including industry leaders such as MiniMax, JD.com, Manycore Tech, DeepRoute.ai, Neolix, Keenon, Independent Variable Robotics, Astribot, RealMan, Noematrix, PingCAP, Zilliz, ASTRONSTONE, and XTransfer.
📩 Contact us
BP submissions or ecosystem partnerships: community@yunqi.vc
WeChat official account: Yunqi Capital (ID: yunqipartners)
Xiaohongshu: Yunqi on Anfu Road
View the transcript of this episode on Xiaoyuzhou