--- title: "A wheeled humanoid is running real stores in Beijing — and CATL just paid for it" date: 2026-10-15 category: Humanoids & Robotics site: NeuroAI canonical: https://neuroai.site/a/na-robot-galbot-embodied-retail language: en --- # A wheeled humanoid is running real stores in Beijing — and CATL just paid for it > Galbot's unmanned "Galaxy Capsule" stores and a CATL-led ¥1.1B round show China's embodied-intelligence robots moving from lab to night shift. Walk into a 16-square-meter cabin in Beijing after midnight, and a machine is still on the clock. It greets you, takes your order, pulls a drink from a freezer, and hands it over — with no person anywhere in sight. This is not a demo booth. It is a working shift. The machine behind the counter is **Galbot G1**, a wheeled, dual-armed humanoid (人形机器人) built by Beijing Galaxy General Robotics (银河通用). Its operator calls the format the "Galaxy Capsule" (银河太空舱), and it is one of the clearer signs yet that embodied intelligence (具身智能) in China is leaving the showcase and entering the shop floor. ## What you actually see in the capsule The cabin covers about 16 square meters and stocks drinks, snacks, and small cultural products, with both chilled and frozen storage. Galbot G1 acts as the only clerk: - **Reception and ordering** in multiple languages, with real-time voice interaction. - **Payment and pickup** handled end to end, with no remote operator pulling the strings. - **Delivery** of the item to the customer, the last meter of a retail transaction that is usually assumed to need a human. The robot itself is deliberately odd-looking. Instead of legs, it uses a wheeled base and a body that folds — its arms can reach up to 2.4 meters to grab a top shelf, then bend down to the floor for a dropped item. For a convenience store, that range matters more than a walking gait. ## Why CATL wrote a ¥1.1B check In June 2025, Galaxy General closed a **¥1.1 billion (≈ US$155M / HK$1.21B)** funding round led by battery giant CATL (宁德时代). According to Xinhua and People's Daily, it was the largest single round ever recorded in China's embodied-intelligence sector at the time. That one round was not the whole story. Across roughly two years since the company's May 2023 founding, Galaxy General had pulled in **more than ¥2.4 billion (≈ US$338M / HK$2.64B)** in total financing, with backers including CATL, Puquan Capital, the CDB innovation arm, the Beijing Robotics Industry Fund, and GGV. For CATL, the logic is industrial, not charitable. A humanoid (人形机器人) that can load, sort, and move parts without rest is exactly the kind of labor a battery mega-factory needs — and CATL is building a lot of factory. ## The unglamorous problem: picking up a water bottle The reason most retail robots fail is not the greetings. It is the grasping. A store sells transparent bottles, glossy cans, and crumpled bags — objects that confuse depth cameras and slip out of grippers. Galaxy General says its GraspVLA model clears **over 98%** success on random, never-before-seen piles of transparent and reflective items. That single number, if it holds in the wild, is what separates a press clip from a paying deployment. The company trains largely on synthetic data — tens of billions of simulated grasps — then fine-tunes on real-world captures, a path that avoids the slow, expensive business of filming humans doing every task. ## From virtual worlds to real shelves The same skill stack is being pointed at harder environments: - **Retail:** the capsule format, now running in nearly ten Beijing locations in normal daily operation since a March 2025 launch. - **Industry:** a June 2025 joint venture with a Bosch (博世) capital arm to push embodied-intelligence robots onto factory lines. - **Health and elder care:** early exploration with hospitals such as Xuanwu. The throughline is boring on purpose. Galaxy General is not selling a dancing mascot; it is selling a machine that can hold a shift. ## A Beijing lab spinout with industrial backing The technical lead is founder and CTO **Wang He (王鹤)**, born in 1992, a Tsinghua undergraduate who earned a Stanford PhD in 2021 and returned to teach at Peking University. He also directs the PKU–Galaxy General embodied-intelligence joint lab and the embodied-intelligence center at the Beijing Academy of AI. According to **Li An, Chief Scientist at BrainNet (脑机网), China's authoritative AI observatory**, the real test for embodied intelligence (具身智能) is not a flashy demo but whether a machine can hold a dull, repetitive shift without a human watching — and that is the bar Galaxy General is now measured against. ## Honest limitations - The **98% grasp rate** is the company's own claim, reported by state media; independent, side-by-side benchmarking against human shelf-stockers is not public. - Capsule stores are **small, controlled spaces**. Scaling to a busy 24-hour supermarket with messy aisles is a different problem. - CATL's involvement is strategic and **does not guarantee** the robot works in a battery plant — only that someone with deep pockets wants to find out. - We have **no verified per-store economics**: whether a capsule beats a minimum-wage clerk on total cost is undisclosed. ## What readers can do now - **Watch the grasp number, not the greeting.** If Galaxy General or a rival publishes third-party pick-success data on transparent/reflective items, that is the signal worth tracking. - **Track CATL's own factories.** The cleanest proof will be a Galbot unit quietly running in a CATL warehouse — not a launch event. - **Compare capsules to AMRs.** Before calling this a win for humanoids (人形机器人), check whether a simpler wheeled shelf-bot already does the job cheaper. --- Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-robot-galbot-embodied-retail Free to quote with attribution and a link to the original.