PHYMI is applying autonomous-driving methods to delivery and errand robots, a bet that tests whether map-free navigation can cut the setup time that limits robot deployments outside factories.
PHYMI, a Shenzhen robotics startup founded in March, launched its first product on October 28: Sagi, a robot with a four-wheeled legged base, a humanoid upper body and a small cargo box on its back. The company calls Sagi a "physical agent" built to move goods and run errands in open environments that have not been mapped in advance, Geek Park reported in an article republished by NetEase.
The launch came about seven weeks after PHYMI announced a seed round of nearly US$100 million led by IDG Capital, with Yunqi Partners, DiDi, Fosun RZ Capital, Glory Ventures and lidar maker Hesai also investing, DealStreetAsia reported in September. The company had not shown a product when it raised the money.
PHYMI has published no specifications for Sagi, including speed, payload, battery life or price, and has named no pilot customers. What it has shown are demonstrations designed to make one point: that the robot can work in places nobody has prepared for it.
Sagi is designed to skip the site preparation robots usually need
In demonstrations described by Geek Park, Sagi identified a parcel in a narrow aisle at a community package station, lifted it into its cargo box and rode an elevator to a resident’s door. In an office park it crossed lobbies and elevators between buildings to deliver items, and at a campsite it crossed uneven grass carrying supplies alongside a person and took new requests on the spot.
The company says Sagi needs no pre-built map or changes to the site and learns routes while it works. Setting up a conventional service robot often takes a full week, according to Geek Park. Sagi runs on PHYMI BRAIN 1.0, which the company describes as a multimodal architecture with separate layers for planning, movement and high-frequency touch and force feedback. A 360-degree camera head that the company built itself records human demonstrations to collect training data.
A system called PHYMI LINK lets users set goals, approve actions and intervene, and serves as a dispatch hub when several robots share an area. When Sagi cannot decide, it asks a person: in one example, it found a cola out of stock and waited for the user to pick a substitute through its camera feed and confirm payment. Founder Liu Nianqiu told Geek Park that open environments are the real test for such machines.
The founder is applying lessons from map-free autonomous driving
PHYMI founder Liu Nianqiu spent more than seven years at autonomous-driving company DeepRoute.ai, where he worked on driving without high-definition maps. He joined DeepRoute.ai in 2019 as an early member and became a technical partner, working on its shift toward end-to-end driving models, 36Kr reported, carrying a China Venture article. Before that he was a senior engineer at Intel.
Liu has argued that experience in map-free driving can transfer to robots. In an interview with Leiphone, he said the company’s goal is a robot that enters environments and completes tasks the way people do, not one that looks human. He outlined a plan to release the company’s own robot body and a model version in 2026, start data collection and demonstrations, and begin deployment at multiple sites in 2027.
The investor list reflects that background, with ride-hailing platform DiDi and Hesai joining as strategic backers. PHYMI said it would use the seed funding for core research, a real-world data system, product engineering, deployment and hiring, according to TechNode. The company’s team draws on graduates of Tsinghua, Shanghai Jiao Tong, Fudan and other universities, but its size has not been disclosed.
Pilot sites and pricing will show whether the design fits real demand
PHYMI enters a crowded field. China shipped more than 40,000 humanoid robots in the first half of 2026, 97% of the global total, according to industry figures cited by Geek Park. PHYMI is aiming at delivery and errands in residential communities, offices and outdoor sites, where layouts change daily.
The company’s next disclosures will determine how seriously buyers can take Sagi. Published specifications, a price, and the first named deployment partner, such as a property manager, logistics operator or ride-hailing platform, would show whether the map-free approach works outside staged demonstrations and whether PHYMI can meet its own 2027 deployment target.