--- title: "A Shenzhen Lab Built a Bipedal Robot You Can Buy — and Alibaba Took a Stake" date: 2026-10-08 category: Humanoids & Robotics site: NeuroAI canonical: https://neuroai.site/a/na-robot-limx-dynamics-humanoid language: en --- # A Shenzhen Lab Built a Bipedal Robot You Can Buy — and Alibaba Took a Stake > LimX Dynamics, a Shenzhen startup founded in 2022, sells a multi-form bipedal research robot worldwide and has drawn Alibaba and other majors into its funding — a different path from the humanoid hype cycle. A robotics PhD student in Munich unboxes a two-legged machine, swaps its feet from wheels to point-tips in a few minutes, and watches it walk. The robot was not built in Boston or Tokyo. It shipped from a Shenzhen startup that, three years earlier, barely existed — and whose backers now include some of China's heaviest corporate names. LimX Dynamics (逐际动力) was founded in January 2022 by *Zhang Wei (张巍)*, a professor at the Southern University of Science and Technology (南方科技大学) and former associate professor at Ohio State University. The company's bet is unusual: instead of chasing a headline humanoid, it first shipped a *multi-form bipedal robot (多形态双足机器人)* aimed at researchers, then used that revenue and attention to fund a push into full humanoids. ## One machine, three ways to walk The product is called *TRON 1*. Launched for sale in October 2024 with simultaneous release in China and overseas, its pre-order price started at **RMB 69,800 (about $9,800 / HK$76,100)** — far below the high-five-figure sums full-size humanoid development platforms command. Its signature feature is a *"three-in-one" modular foot-end* design: a point-foot for simple control, a sole for humanoid-like walking, and a wheeled foot for efficient all-terrain travel. Users swap ends quickly, and the robot auto-detects the mode. This is deliberately a research platform, not a home appliance. LimX ships an open SDK, supports Python for full development, and offers one-click *Sim2Real* deployment so reinforcement-learning algorithms validated in simulation (NVIDIA Isaac, Mujoco, Gazebo) can run on the real machine. The pitch to universities and labs is simple: one physical body, many locomotion experiments, low barrier to entry. ## Why researchers, not households, are the first buyers Most bipeds announced today are enterprise demos or lab prototypes. TRON 1's distinction is that it was *sold*, with reservations opened to developers, universities and commercial R&D teams. That matters because the hardest unsolved problem in robotics is not building one impressive machine — it is building a machine whose locomotion and control generalize across floors, curbs, stairs and clutter. LimX pairs the hardware with algorithm work. It has published *LimX VGM*, an embodied-manipulation method that uses video generation to improve how a robot's "brain" learns, and it emphasizes reinforcement-learning-based *whole-body motion control (全身运动控制)* — the "cerebellum" that keeps a legged machine upright and efficient. The company says its three core threads are hardware design and manufacturing, RL-based cerebellar control, and embodied-brain model training. ## The Alibaba vote of confidence In March 2025, LimX announced an *A+ round* that brought its A-series financing to **RMB 500 million (about $70 million / HK$545 million)** within half a year. The investor list reads like a who's-who of Chinese strategic capital: *Alibaba*, *China Merchants Venture (招商局创投)*, *Shangqi Capital (尚颀资本, tied to SAIC)*, *NIO Capital (蔚来资本)* and *Lenovo Capital*, among others. Earlier, a 2023 angel and Pre-A round totaled roughly **RMB 200 million (around $28 million / HK$218 million)**, and a July 2024 A round was led by Alibaba and China Merchants — notably Alibaba's first known investment in a humanoid-robot company. For a startup, that backing is a signal about *where* the money thinks locomotion and embodied AI will pay off: not necessarily in a humanoid that folds your laundry, but in platforms and toolchains that let many developers teach machines to move and act. ## From a walking platform to a full humanoid LimX has not stopped at the research biped. Through 2025 and 2026 it iterated full-size humanoids and, in May 2026, unveiled *LimX Luna*, a full-size interactive humanoid, alongside a "brain" system called *COSA* (updated to version 0.5 in July 2026). In July 2026 the company also disclosed a *Pre-IPO round*. The arc is clear: prove locomotion and a developer ecosystem with TRON 1, then climb toward a general-purpose humanoid. That sequencing is the opposite of the typical hype curve. Rather than reveal a polished humanoid and hope for orders, LimX monetized the hard middle step — reliable, configurable bipedal movement — and let research revenue and strategic capital fund the climb. ## Honest limitations TRON 1's launch (October 2024) and price (from RMB 69,800) are confirmed by the company's own release and by China Securities/China Economic News coverage; this is a 2024 product still being commercialized, not a 2025–2026 debut. The RMB 500 million A-series, the RMB 200 million angel/Pre-A, and the Alibaba-led July 2024 A round are reported by *Securities Times (证券时报)* and corroborating outlets, but remain company-announced funding figures. The Luna humanoid, COSA 0.5, and the July 2026 Pre-IPO round are stated on LimX's own website and were not independently audited here. "World's first multi-form bipedal robot" is the company's marketing claim; similar modular bipeds exist from other labs, so the superlative should be read narrowly as LimX's commercial framing rather than a verified absolute. TRON 1 is explicitly a technical/research product, not a consumer robot, and its real-world autonomy on stairs and clutter is still operator-assisted in many cases. ## What readers can do now - If you follow robotics investing, separate "humanoid reveal" news from "robot actually sold" news — LimX's TRON 1 is in the rarer second category, and that is the more meaningful signal. - University and lab readers can treat sub-$10,000 modular bipeds like TRON 1 as a legitimate entry point for locomotion and Sim2Real research, provided you budget for the SDK learning curve. - Watch which embodied-AI startups attract *strategic* (Alibaba, SAIC, NIO) rather than purely financial backers — that split often predicts who gets real deployment partners. --- Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-robot-limx-dynamics-humanoid Free to quote with attribution and a link to the original.