PHYMI Closes Nearly $100 Million Seed Round to Bring Physical Agents into Open, Dynamic Real-World Environments | Yunqi Partners

When intelligence is no longer confined to conversations on a screen — when it can move, manipulate, and make decisions on its own — the Physical Agent becomes the key vehicle for the next stage of embodied AI.

PHYMI is answering this challenge with a full-stack system: building a Physical Agent that understands human goals and keeps acting in open, dynamic, real-world environments. Founder and CEO Nianqiu Liu previously served as a technical partner and board member at DeepRoute.ai, where he witnessed firsthand China's advanced autonomous driving industry move from L4 R&D to mass production at scale. The team brings together frontier model expertise, full-stack robotics engineering, and mass-production delivery experience.

Recently, PHYMI closed a seed round of nearly US$100 million, with Yunqi Capital participating — another step in our Physical AI strategy.

Yunqi Capital's investment perspective:

AI is moving from "answering questions" on a screen to "achieving goals" in the physical world, and the Physical Agent is a direction worth watching for the long term. Operating in open, dynamic, real-world environments requires an agent that continuously understands human intent, and can move, manipulate, judge, and decide — pushing complex tasks step by step toward a result.

The PHYMI team combines frontier model vision, complex software-hardware engineering, and mass-production experience. They have a systematic view of both the technical roadmap and productization of the Physical Agent, along with strong engineering execution. We're glad to be alongside them from the earliest stage, and we're long-term believers in PHYMI bringing the Physical Agent into the real world — gradually becoming a trusted action partner for everyone.

Recently, PHYMI announced the completion of a seed round of nearly US$100 million. The round was led by IDG Capital, with participation from Yunqi Capital, DiDi, Fosun RZ Capital, Yaotu Capital, and Hesai Technology.

Founded in 2026, the company won support early on from leading financial investors, industrial capital, representative tech companies, and entrepreneurs. This diverse investor lineup also draws attention to the core question PHYMI has long sought to answer: when AI crosses the digital boundary into the physical world, can it truly understand human goals, adapt autonomously to open, dynamic, real-world environments, and drive complex goals to completion?

Focused on the Open, Dynamic, Real World

Building the Physical Agent

As multimodal large models and agentic AI continue to advance, artificial intelligence is reaching a new inflection point in capability. What people expect from intelligence is no longer just "getting an answer" — it's "giving it a goal, and having the intelligence drive it to completion."

AI today is crossing the boundary of the digital world, and "from providing answers to achieving goals" is becoming an important direction for embodied AI. But unlike the digital world, with its unified interfaces and rules, the physical world has no pre-written script: space keeps extending, environments and human needs change at any moment, and the outcome of one action can alter the path ahead. Machines no longer face a preset script — they face a goal that must be continuously understood and pursued amid change.

PHYMI was built to bridge this gap. PHYMI does not define capability by single-point actions at fixed workstations, in single scenarios, or within preset workflows. Instead, it is committed to building a Physical Agent for the open, dynamic, real world.

"Real" means the machine must directly confront physical laws, errors, and unexpected disturbances. "Open" means spaces, objects, and tasks cannot be enumerated in advance. "Dynamic" means the environment and needs keep changing throughout the course of action.

This is the world each of us lives in every day. Today, AI can process extremely complex information, yet still struggles to take on the real-world tasks that require physical presence and consume time and attention: picking up and delivering items, buying something needed on short notice, restoring order to a space that's been used, or completing multiple tasks in sequence over the course of a trip. These needs seem trivial, but they follow no uniform process, and conditions on the ground and task objectives can change at any time. To take them on, a machine can't just know one action, nor can it be confined to a fixed workstation, a single scenario, or a preset workflow. It must truly understand human goals, keep moving, manipulating, and judging across different spaces, and organize scattered task steps into a complete result.

That's why PHYMI's core focus is whether a Physical Agent can, in the open, dynamic, real world, connect cross-scenario movement and manipulation, keep acting as conditions change on the ground, and drive complex goals to completion.

Intelligence in the digital world can demonstrate capability through answers; intelligence in the physical world must prove its value through action and results.

From Executing Actions to Achieving Goals

Based on this conviction, PHYMI focuses on long-horizon, complex movement and manipulation in the open, dynamic, real world.

"Long-horizon" here means not just greater travel distance, but longer task durations, more action steps, larger spatial spans, and more continuous chains of decision-making. In PHYMI's technical vision, a Physical Agent's capabilities evolve level by level along "action → task → goal": actions are concrete, short-range, and deterministic; tasks consist of multiple actions with a clear process; goals are more open and long-horizon, requiring the machine to continuously handle environmental change and uncertainty.

Moving from "executing actions" to "achieving goals" therefore tests not just locomotion and manipulation, but goal understanding, task decomposition, reasoning and planning, dynamic decision-making, and complex execution.

Like a real-world partner, a Physical Agent must not only understand what a person wants to do — it must, without step-by-step direction, break that goal down into concrete tasks and translate those tasks into a series of physical actions. A person simply states a goal, and the agent maintains task context, autonomously planning and driving the entire process from understanding the need to delivering the result. When the environment changes or new needs arise mid-task, it can respond promptly, adjust the plan, and chain new tasks seamlessly onto the old.

PHYMI wants people to confidently hand goals over to the Physical Agent: the person sets the goal, and it crosses space, organizes actions, and delivers results. When machines truly gain the ability to continuously achieve goals, their relationship with people begins to shift from tool to partner.

Building a Full-Stack Physical Agent System

for the Open, Dynamic, Real World

To bring "goal achievement" into the physical world, you can't simply bolt an intelligent model onto a robot body. Models, hardware, data, and interaction must work together around a single goal. To this end, PHYMI has built a full-stack technical system composed of PHYMI BRAIN, PHYMI BODY, PHYMI DATA, and PHYMI LINK.

• PHYMI BRAIN: for the open, dynamic, real world, an embodied AI model system spanning multimodal perception, world modeling, reasoning and planning, and action generation — enabling the Physical Agent to understand people and the world, predict physical interactions, and autonomously complete movement and manipulation tasks;

• PHYMI BODY: a robot body built for movement and manipulation in the open, dynamic, real world, covering mechanics and actuation, sensing and perception, compute and communications, energy systems, and safety and reliability — giving the Physical Agent a physical vessel that can enter real environments and run stably over the long term;

• PHYMI DATA: accumulating multi-perspective, cross-modal real-world dynamic data and task feedback, turning every interaction into experience that can be accumulated and learned from, continuously driving the evolution of intelligence;

• PHYMI LINK: connecting people, the Physical Agent, and the environment, so that goals, authorization, status, and feedback run through the entire task lifecycle — and supporting human-machine and machine-machine collaboration.

Together, these four systems form a complete closed loop: BRAIN understands the world and decides on actions, BODY provides the physical vessel for entering the real world and acting continuously, DATA turns interactions into experience and drives the evolution of intelligence, and LINK runs through the entire task lifecycle.

Through this full-stack system, PHYMI aims to make the Physical Agent capable not just of executing actions, but of continuously acting, completing tasks, and ultimately achieving human goals in the open, dynamic, real world.

A Hybrid Team Spanning Frontier Models

to Mass Production at Scale

Bringing frontier AI into the open, dynamic, real world requires not just model capability, but coordination across complete robot systems, data loops, and engineering at scale. PHYMI's team strength lies in connecting the full chain from frontier research to real products.

PHYMI founder and CEO Nianqiu Liu previously served as technical partner and board member at DeepRoute.ai, where he witnessed firsthand China's advanced autonomous driving industry move from L4 R&D and operations to mass production at scale — bringing both forward-looking judgment on next-generation technology directions and the ability to organize teams to turn technical strategy into products at scale. At DeepRoute.ai, Liu participated in the deployment of map-free autonomous driving solutions and the evolution of the technical stack toward end-to-end models and VLA large models. Earlier, he was a senior engineer at Intel and has serial entrepreneurial experience in smart hardware.

From technical judgment in open-world problems, to complex software-hardware engineering, to supply chain, mass-production delivery, and real-world operations — this complete set of experience forms an important foundation for PHYMI's push to bring the Physical Agent into the real world.

PHYMI's core members come from Tsinghua University, Shanghai Jiao Tong University, Fudan University, Nanyang Technological University, The Chinese University of Hong Kong, and other universities. Team members have served as core technical and product leads at leading internet, humanoid robotics, and autonomous driving companies, and have led work on large models, VLA, real-robot reinforcement learning, complete robot systems, data platforms, and mass-production delivery.

This is a hybrid team combining frontier model research, full-stack robotics engineering, mass production at scale, and data-loop capability — one that can connect models, data, hardware, and real-world validation, turning frontier research into a Physical Agent that runs, iterates, and scales.

Becoming Everyone's Action Partner

This round of funding will primarily go toward core Physical Agent technology R&D, building the real-world dynamic data system, product engineering and scenario deployment, and team expansion — accelerating the Physical Agent's product iteration and development.

PHYMI wants the Physical Agent to enter the daily lives of ordinary people, taking on the high-frequency, trivial tasks that still require someone to do them by hand. It doesn't need to be omnipotent, and it doesn't need to be a "superman." It should be reliable, practical, and affordable — able to lend a hand by your side, and able to act independently with authorization; able to keep handling things when circumstances change, and knowing when to ask a person for help when it's unsure.

As shared experience accumulates, it will grow increasingly familiar with its surroundings and increasingly attuned to people's habits and needs — becoming an action partner that shares goals with people and gets things done together.

Let intelligence step out of the screen and into the open, dynamic, real world.

Crossing space, achieving goals, becoming everyone's partner.

GO ANYWHERE. DO ANYTHING. WITH ANYONE.