Liu Nianqiu, a veteran in the intelligent driving industry, has founded his own startup and secured nearly 100 million US dollars in the first round of financing.

On September 9, ChinaVenture learned that PHYMI, an embodied AI startup, has closed a nearly 100 million US dollar seed round financing. This round is led by IDG Capital, with participation from Yunqi Capital, DiDi, Fosun RZ Capital, YoToo Capital, and Hesai Technology.

At present, competition in the embodied AI industry has reached a white-hot stage, with more than 20 players boasting a valuation of over 10 billion yuan. As a newly established startup still in its early stage, PHYMI has demonstrated outstanding fundraising appeal in this round, both in terms of financing amount and investor lineup.

One of the core anchors for capital betting lies in the team. Liu Nianqiu, Founder and CEO of the company, previously served as Technology Partner and Director of DeepRoute.ai, and personally experienced the key stage when China's high-level autonomous driving developed from L4 R&D and operation to large-scale mass production. Most of its core members also come from leading AI internet, humanoid robot and autonomous driving enterprises, who have led the work of large models, VLA, real machine reinforcement learning, robot complete machine, data platform and mass production delivery.

It can be said that this is exactly the embodied team profile most valued by the current capital market, which integrates model vision, complex software and hardware engineering capabilities, and large-scale mass production and implementation experience.

From the perspective of the track, with the continuous breakthroughs of multimodal large models and physical AI, the market's expectation for embodied intelligence is undergoing qualitative changes: it is no longer satisfied with the display-type capabilities of "being able to run, jump and perform", but requires robots to truly enter the physical world, understand and independently complete the goals set by humans.

PHYMI's core direction is exactly at the node of this demand upgrade. It is positioned to build a Physical Agent oriented to the open, dynamic and real world. In simple terms, it enables robots to understand human goals, move and operate autonomously across scenarios, and continuously complete real tasks, so as to liberate humans from high-frequency, trivial affairs that always require personal handling.

Regarding the use of financing, Liu Nianqiu said that the funds will be mainly used for the R&D of core Physical Agent technologies, the construction of real dynamic data systems, product engineering and scenario implementation, as well as team expansion.

Another veteran in intelligent driving devotes to embodied intelligence

The story of PHYMI starts with the soul of the company, its founder Liu Nianqiu.

Before founding PHYMI, Liu Nianqiu's resume was already full of hardcore experience. As early as when he was a senior engineer at Intel, he was engaged in R&D and implementation of new equipment, new processes and automation, and accumulated rich experience in the engineering implementation of complex software and hardware systems. After leaving Intel, he participated in a number of intelligent hardware projects, further accumulating experience in product R&D and continuous entrepreneurship.

In 2019, Liu Nianqiu joined DeepRoute.ai, which was still in its early startup stage at that time, and became a technology partner. As a veteran of the company, he participated in promoting the implementation of the company's map-free intelligent driving solution, and the evolution of the technical system to end-to-end models and VLA large models.

It is these experiences around intelligent hardware that made Liu Nianqiu see the potential of the robot field very early, but he did not rush to start his business. In his view, the wave of robot entrepreneurship that emerged around 2024 was still in the development stage of robot body supply chain, motion control, data and model paradigm, which essentially relied on traditional algorithms and optimization theories, and was far from reaching the AI-driven "intelligent" level.

The deeper restriction lies in the hardware side — at that time, there were almost no joint motors and special reducers specially designed for robots on the market. Traditional servo motors in the industrial field were difficult to be directly installed into robots due to their large size and insufficient unit torque density. This means that if you want to do embodied intelligence at that node, you need to invest a lot of energy to start from the basic components.

"Although hardware is very important for robots, I believe that a single hardware capability is not the core variable that determines the success or failure of future embodied intelligence. What I want to make is a systematic product, and the core of the continuous evolution of the product is its 'brain' capability." Liu Nianqiu said.

For this reason, he chose to wait, until 2026, the timing finally came. Liu Nianqiu told me that this judgment is mainly based on three levels of changes:

The first is the maturity of the supply chain. Nowadays, there are a large number of motors and reducers specially optimized for robots on the market, which greatly reduces the difficulty of body manufacturing. Secondly, the supply of AI infrastructure is far more abundant than it was two years ago. More importantly, in the past two years, the intelligent driving field has fully completed the large-scale data closed loop of "data collection — cleaning and labeling — model training — simulation verification — OTA iteration", and this whole set of methodology has high reference value, of course, the specific method cannot be copied directly.

"This means that we don't need to spend a year or two 'reinventing the wheel', but from the first day, we can focus on thinking about the product itself, and building a closed loop for the continuous iteration of products, data and models."

So in March 2026, PHYMI Technology was officially established.

Building a Physical Agent for the Open, Dynamic and Real World

The English name of PHYMI is "PHYMI", where "PHY" stands for PHYsical World, "M" corresponds not only to Mobility, but also to Manipulation, and "I" points to Intelligence.

Liu Nianqiu said that this is also the core problem the company wants to solve: to build a Physical Agent oriented to the open, dynamic and real world, focusing on long-distance and complex movement and operation. He further explained that the "long-distance" here not only means a longer movement distance, but also includes longer task time, more action steps, larger space span, and a more continuous decision-making chain.

Behind this choice is the team's judgment on the evolution direction of embodied intelligence. At present, with the continuous development of multimodal large models and Agentic AI, AI is crossing the boundary of the digital world — people's expectation for intelligence is no longer just "getting an answer", but "giving a goal, and the intelligence will independently promote the achievement of the goal".

However, the physical world is far more complex than the digital world. The digital world has unified interfaces and rules, but the physical world has no pre-written script: the space is expanding, the environment and people's needs are changing at any time, and the result of one action may also change the subsequent path.

Take the scenario of "picking up takeout" as an example. On the surface, it is just a simple task of two points and one line, but in the real scenario, the position of the item, the delivery method and the on-site state may all change. This means that the robot must not only understand human intentions, but also break down the goal into specific tasks without step-by-step command, and then convert the tasks into a series of physical actions. When the environment changes or new demands are added temporarily, it can also respond in time, adjust the plan, until the task is completed.

In reality, the capabilities of a large number of current robots are still concentrated in fixed stations, fixed processes or specific scenarios, and do not have such flexible response capabilities.

This is exactly the gap that PHYMI wants to fill. Therefore, from day one, the company does not define capabilities by single-point actions in fixed stations, single scenarios or preset processes, but focuses on whether Physical Agent can connect cross-scenario movement and operations, and continue to act according to on-site changes to promote the achievement of complex goals.

However, to make the capability of "achieving goals" truly enter the physical world, it is obviously not enough to simply superimpose the intelligent model and the robot body. It requires the model, body, data and interaction to work together around the same goal. To this end, PHYMI has built a full-stack technology system consisting of PHYMI BRAIN, PHYMI BODY, PHYMI DATA and PHYMI LINK.

BRAIN is responsible for understanding the world and deciding actions, BODY provides the physical carrier to enter the real world and act continuously, DATA transforms every interaction into accumulable and learnable experience by precipitating multi-view, cross-modal real dynamic data and task feedback, and finally LINK connects people, Physical Agent and the environment, so that goals, authorization, status and feedback run through the whole process of the task. The four systems are interlocked to form a complete closed loop starting from the goal, covering perception, decision-making, action, feedback and evolution.

Although from the perspective of product positioning, PHYMI is targeting cross-scenario general capabilities in the long run, Liu Nianqiu does not plan to shape Physical Agent into an all-powerful "superman", let alone an automated machine designed to replace humans.

In his vision, PHYMI's Physical Agent should be a reliable, practical and affordable "action partner" around everyone — in the future, it will enter the daily life of every ordinary person, undertaking high-frequency, trivial real affairs including picking up and delivering items that always require someone to be present in person and continuously consume time and attention.

"It can not only help people around, but also take charge independently after getting authorization; it can continue to move forward when things go smoothly, adjust flexibly when encountering changes, and know how to ask for help in time when it is not sure. And with the continuous accumulation of common experiences, it will become more and more familiar with the surrounding environment, more and more understand people's habits and needs, and finally be able to share goals with people and accomplish things together." Liu Nianqiu said.

This article is from the WeChat official account "ChinaVenture", Author: Wang Manhua, Editor: Wang Qingwu, 36Kr is published with authorization.