Key takeaways
- Galbot (银河通用), founded in May 2023 in Beijing by PKU professor Wang He (王鹤), builds the G1 humanoid (人形机器人) for retail, industrial, and care scenarios.
- Its latest funding round was ≈ US$155 million / HK$1.20 billion in June 2025, led by CATL (宁德时代); cumulative funding exceeds ≈ US$338 million / HK$2.62 billion.
- The G1, launched in June 2024, uses a wheeled base plus dual arms and a foldable single leg; it stands 1.73 m with a 190 cm arm span and a 10-hour battery.
- Rather than chase factories first, Galbot started with pharmacies and stores, where tasks are repeatable and return on investment is legible.
- At the 2025 Zhongguancun Forum the G1 served coffee thousands of times over five days at a success rate above 99.97%.
- Embodied-AI models GraspVLA, GroceryVLA, and TrackVLA train on synthetic data first, then refine on real deployments.
The pharmacy that never sleeps
Step into one of a handful of Beijing pharmacies after dark and there may be no clerk behind the counter — only a humanoid (人形机器人) reading the shelf, fetching a box, and handing it over. It is not a stunt. The machine works unattended, processing more than 300 orders a day at some locations. The company behind it, Galbot (银河通用), made a deliberate choice to start with the dullest, most structured retail job it could find, and that restraint is the interesting part.
Galbot was founded in May 2023 in Beijing by Wang He (王鹤), a Tsinghua undergraduate, Stanford PhD, and now an assistant professor at Peking University who directs the university's embodied-intelligence joint lab with the company.
The money and the model
Funding is the best-documented part. Xinhua reported on 23 June 2025 that Galbot closed a ≈ US$155 million / HK$1.20 billion round led by CATL (宁德时代), with Guokai Science and Technology Innovation, the Beijing Robotics Industry Fund, and GGV participating. Combined with a ≈ US$99 million / HK$765 million angel round in July 2024 and a ≈ US$70 million / HK$546 million strategic round in November 2024, cumulative funding exceeds ≈ US$338 million / HK$2.62 billion — described by the report as the largest single-category raise in China's embodied-AI field to that point.
The product is the G1, launched in June 2024: a wheeled base with dual arms and a foldable single leg, 1.73 m tall, 190 cm arm span, 10-hour battery. Its software rests on a clear recipe — train on synthetic, Sim2Real data first, then refine on real deployments — expressed in models named GraspVLA (grasping), GroceryVLA (grocery handling), and TrackVLA (navigation).
Why retail before the factory floor
Factories are the headline ambition for most humanoids, but Galbot went the other way first. A pharmacy has few stock-keeping units, standardized packaging, and a task — find, pick, hand over — that repeats thousands of times. That is exactly where a robot's success rate is easy to measure and where a business case is easy to defend. By March 2025 the company had deployed its smart-retail solution in nearly ten Beijing stores for normal daily operation; by June 2025, seven unmanned pharmacies were running in the city.
The proof point was the 2025 Zhongguancun Forum, where the G1 served coffee thousands of times across five days at a success rate above 99.97% and was named an official "steel model-worker volunteer." Its smart-pharmacy solution was selected among the forum's top ten scientific achievements. In June 2025 Galbot also formed a joint venture with Bosch's investment arm to explore industrial scenarios — the factory pivot arriving only after retail had been de-risked.
Why a foldable leg instead of two feet
Galbot's G1 is neither a pure wheeled box nor a bipedal athlete. It pairs a wheeled base with a single foldable leg and dual arms, letting the body rise from 0.8 m to 2.4 m to reach high shelves without the cost and instability of full bipedal walking. The design encodes a pragmatic view of embodiment: most retail and pharmacy tasks happen between knee and head height, and a controlled vertical lift covers that range more reliably than legs. The trade-off is obvious — it cannot walk through a cluttered stockroom — but for a machine whose job is to stand, reach, and hand over, that is a reasonable sacrifice.
Honest limitations
- The 99.97% success rate was measured in a curated forum demo, not in a full-year commercial pharmacy deployment; real-world rates will vary with crowding and edge cases.
- "Thousands of orders a day" at pharmacies describes early pilot sites, not a mature national network; scale and unit economics are still being proven.
- Heavy strategic backing from CATL signals industrial alignment but also concentrates dependence on a few large partners.
- Synthetic-first training helps with generalization, yet truly unstructured retail shelves remain a known hard problem.
What readers can do now
- Retail and pharmacy operators in China should track Galbot's Beijing pilot expansion as a live test of unattended humanoid economics, not the keynote clips.
- Observers should watch whether the Bosch joint venture converts into paid industrial deployments — that is the real signal that the retail-first bet paid off.