Robot Forward reported on 9 September that Feimo Yi Technology, operating as PHYMI, has closed a near-US$100 million seed round. The round was led by IDG Capital, with Yunqi Capital, Didi Chuxing, Fosun Ruiheng, Yaotu Capital and the lidar maker Hesai Technology joining as follow-on investors.
From autonomous driving to the physical world
The company sits behind founder Liu Nianqiu, a serial entrepreneur who moved from autonomous driving into embodied intelligence. He was formerly a senior engineer at Intel and later a technical partner and director at Yuanrong Jiexing, where he helped ship mapless driving and steered the technology stack toward end-to-end and VLA large models.
At Yuanrong he watched self-driving travel from zero to one, then from one to ten and on to mass production, and saw China’s high-level autonomous driving mature from L4 research into scaled manufacturing. Robotics, in his reading, only entered its true zero-to-one phase on the back of AI from 2025, as foundation models, world models and physical-intelligence methods began to converge.
The name PHYMI stands for Physical World, Manipulation and Mobility, and Intelligence. The Chinese name is a phonetic rendering chosen when clearer meanings proved impossible to register, and it quietly nods to Moravec’s paradox, the gap between machine reasoning and real-world motor skill.
A Physical Agent for open, dynamic life
PHYMI wants to build a Physical Agent that steps into open, dynamic living spaces rather than a fixed, fenced-off demo zone. Liu sketches a robot that joins daily life as a helper: at a campsite it carries gear, lays out a mat and fetches coffee; on a walk past an ice-cream shop a user can add a purchase to its task list on the fly. The robot keeps rewriting its own plan while it moves, perceives, manipulates and talks to people.
The pitch that separates PHYMI from the crowd of robot storytellers is zero-threshold deployment, what Liu sums up as GO ANYWHERE, DO ANYTHING, WITH ANYONE. He compares the route to Yuanrong’s early mapless-driving bet: accept that open-space capability is immature at first, then widen the envelope over time.
Around that goal PHYMI is building four layers at once. BRAIN covers multimodal perception, world modelling, reasoning and action for the real world. BODY is the physical carrier that lives and runs for long stretches in real environments. DATA turns real interaction into learnable experience. LINK connects people, the agent and the environment, handling remote interaction, task scheduling and multi-robot coordination.
Why a seed round before the product exists
The near-US$100 million will flow along that system: core Physical Agent research, a real dynamic data system, product engineering, scenario landing and team growth. Liu argues the real test of a data flywheel is whether each turn lifts the product a layer, and that needs genuine market feedback rather than lab scores.
He is blunt that public expectations of robots are inflated, much as early self-driving hype outran the actual brains available. True mass production, in his words, arrives only when ordinary users feel broad, sustained demand and choose to live with a robot daily, not when they buy it for its research or show value.
On architecture, PHYMI uses a multi-layer asynchronous multimodal design rather than a simple slow-thinking-versus-fast-execution split, so different capabilities run at different frequencies and draw on different inputs. Full-modal perception matters: a robot should know its arm has made contact even when the touch lands away from a fingertip sensor.
Editor’s note: This is an adapted translation of the original Zhidx report. It has been trimmed and restructured for readability for an international business audience.
Translated and adapted from Zhidx (https://www.zhidx.com/p/592128.html).