Meta Builds Max Humanoid With China's Fourier, Starts Work on Second Prototype

A robot built in three months, with a successor already underway

Meta Platforms has reportedly built a humanoid robot measuring about 1.5 metres and weighing 48 kilograms with Shanghai robotics company Fourier, completing the prototype in roughly three months before starting work on a second version in September 2026. Internally named Max, the machine combines Nvidia computing hardware, a face that displays expressions and hands designed to pick up small objects.

Li-Chen Miller, Meta's vice president of product for robotics, spent almost two months at Fourier's Shanghai factory during development. Meta's robotics team also travelled to China, and the prototype was demonstrated to some employees at an internal robotics summit in late August.

Meta's longer ambition is to give artificial intelligence a physical presence that could eventually help with household chores. Chief technology officer Andrew Bosworth has reportedly discussed humanoids internally as a possible physical embodiment of Muse, the company's consumer AI agent.

Meta is keeping the current project within its research programme. A company spokesperson described the purpose of the collaboration and its position on selling the machines:

this effort with Fourier is focused on internal research and experimentation, and we have no plans to sell these units.

The combination of an existing robotics platform, engineers working together in Shanghai and a second prototype already in development explains the pace of the effort. Meta is using Fourier's experience to build experimental machines while developing the software that might eventually make them useful around people.

What Max looks like and what its hands can handle

Max stands about 1.5 metres tall, or 4 feet 11 inches, and weighs approximately 48 kilograms, equivalent to about 106 pounds. Its face is a display capable of mimicking expressions, while two speakers mounted in its chest provide sound.

Five fingers on each hand allow the prototype to grasp small objects, reportedly including bottles and toys. Cameras and other sensing hardware help provide information about its surroundings. Nvidia's Jetson Thor platform supplies its computing power.

Fourier's first three generations of humanoid robots reportedly contributed hardware and engineering experience to Max. The companies adjusted the machine's height, proportions, head, display, cameras, hands and shoes to suit Meta's requirements.

Max therefore has its own dimensions and configuration. Fourier's GR-3, for comparison, stands approximately 1.65 metres tall and weighs about 71 kilograms. Using those figures, Max is about 15 centimetres shorter and 23 kilograms lighter, a meaningful difference for a machine intended to interact with people indoors.

Meta wants the robot to show when it is uncertain

Meta's reported design requirements focus heavily on communication. The company wants Max to recognise spoken commands from about five metres away, interpret gestures from around three metres and maintain eye contact during interactions. Those distances are development targets; public performance measurements have yet to be released.

The robot is also intended to signal whether it is listening, processing information, speaking or uncertain about a task. A person asking it to move an object would therefore receive some indication of what stage the machine had reached, rather than having to interpret silence.

Eye contact and facial expressions serve a practical purpose in that design. They can help a person understand whether the robot is responding to them, while an uncertainty signal could communicate that an instruction needs clarification.

For household tasks, understanding a request and communicating progress are closely connected. A robot asked to fetch a bottle needs to identify the right object, work out where it is and make its response understandable to the person waiting for it. Meta's requirements bring those problems into the same research programme.

Max 2 targets better hands and movement on stairs

Development of Max 2 reportedly began in September, shortly after the August demonstration. Planned changes include a redesigned torso, improved hands, a projector and additional sensors intended to help the machine navigate stairs.

Hands and stairs represent two concrete limits on the usefulness of a household robot. Hands determine which objects it can handle and how precisely it can move them. Stair navigation determines whether it can reach another floor or negotiate changes in height within a building.

The projector would add another way to display content beyond the screen on the robot's face. Its intended everyday uses have not been detailed publicly.

Meta's development sequence has moved quickly through several stages:

• February 2025: Meta reportedly formed a dedicated robotics product team, with Marc Whitten appointed to lead the effort.
• May 2026: Meta acquired Assured Robot Intelligence, adding expertise in AI models for robots.
• Late August 2026: Max was demonstrated to some Meta employees at an internal robotics summit.
• September 2026: Work reportedly began on Max 2.
• October 31, 2026: Details of Max and the Fourier collaboration became public through reporting.

Why Fourier offered a fast starting point

Fourier began in Shanghai in 2015, building rehabilitation technology such as exoskeletons and systems used for physical training. That work required expertise in controlling movement, sensing forces and designing machines that operate close to people.

The company began its humanoid project in 2019. An early prototype stood and walked without a tether in 2022, followed by the GR-1 in 2023 and GR-2 in 2024. Fourier introduced the open source N1 and the GR-3 in 2025.

Fourier's actuator technology is part of that foundation. An actuator is the mechanism that turns a control instruction into movement at a joint. Fourier combines motors, drivers and controllers in these units, helping its robots manage motion and respond to physical forces.

The GR-3 is designed around hearing, vision and touch, with up to 55 degrees of freedom. That term describes the independent movements available across a robot's joints and mechanisms. More independent movements give engineers more options for positioning limbs and hands, although useful behaviour still requires software to coordinate them.

Meta could build on those existing systems while customising Max for its own experiments. The three month development period reflects a project with an established mechanical starting point and engineers experienced in assembling and adapting humanoid machines.

A partner with customers, investors and production experience

Fourier brings experience supplying equipment as well as developing prototypes. By August 2025, the company said its rehabilitation products served more than 3,000 institutions and hospitals across more than 60 countries and regions.

Market research firm Omdia estimated that Fourier shipped 300 humanoid robots in 2025, placing it sixth globally. Shipments give a measure of its manufacturing activity alongside the rehabilitation business that preceded its humanoid programme.

Investors have funded that expansion over several rounds. A financing round led by SoftBank Vision Fund 2 raised RMB 400 million in 2022, with participation from Aramco backed Prosperity7 Ventures and Vision Plus Capital. Fourier's Series E financing reportedly reached nearly RMB 800 million in early 2025.

Meta has also reportedly recruited robotics engineers in Shanghai and Shenzhen for work ranging from prototype development to preparations for manufacturing at scale. Those roles extend the company's engineering presence near Chinese robotics suppliers, while its stated purpose for the Fourier machines remains internal experimentation.

What Nvidia's chip contributes

Max reportedly uses Nvidia's Jetson Thor, a computing platform based on the company's Blackwell architecture and designed for robots and other autonomous machines. It processes information from sensors and runs AI models on the device itself.

A robot needs to connect what it senses with what it does. Cameras can supply images of an object, for example, while computing systems help identify it, estimate its position and plan how an arm should move towards it.

Processing information directly on a robot can reduce the need to send every decision through a remote server. That matters for physical actions, where a delayed response can affect how a machine moves around an obstacle or handles an object.

The Jetson AGX Thor platform includes configurations with 128 gigabytes of memory and a Blackwell graphics processor with 2,560 CUDA cores, which are processing units used for parallel calculations. Max's exact computing configuration and the AI models running on it have not been disclosed.

Nvidia's robotics ecosystem also connects hardware with development tools. Fourier was named among humanoid developers working with Nvidia's Project GR00T in 2024, giving the companies an existing point of contact around AI for physical machines.

How Muse and the robotics acquisition fit together

Bosworth's reported description of humanoids as a physical embodiment of Muse connects the robot programme to Meta's broader AI ambitions. An AI agent is software designed to take steps towards a goal on a user's behalf, rather than simply produce an answer to a question.

A physical version would need to turn instructions into movements. Fetching an item from another room, for example, requires locating the object, planning a route, grasping it and returning it safely. Language understanding becomes one part of a longer chain of actions.

Meta's May acquisition of Assured Robot Intelligence adds expertise aimed at that connection between intelligence and movement. Founded by robotics researchers Lerrel Pinto and Xiaolong Wang, the startup develops AI systems intended to help robots understand and adapt to human behaviour.

The arrangement gives Meta two complementary resources: Fourier's mechanical engineering and robot platforms, and additional internal expertise in robot learning and control. The long term goal is a machine whose software can respond to changing situations while its body carries out useful tasks.

What households would need from a future version

Max's reported bottle and toy handling provides a concrete starting task. A useful household assistant would need to perform actions repeatedly across different objects, rooms and arrangements, including situations where someone moves furniture or interrupts an instruction.

A machine weighing 48 kilograms would also need dependable movement around people and possessions. Object handling, balance, sensing and clear communication would all affect whether someone could comfortably share a room with it.

Meta's planned changes for Max 2 address parts of that practical challenge: improved hands for manipulation, more sensors for movement and a revised body. The reported communication requirements address another part, helping people understand what the machine is doing.

Commercial decisions lie beyond the current research effort. Meta has announced no consumer launch date, price or subscription plan for Max or Max 2. For now, the concrete development is an experimental robot built with Fourier, followed by work on a more capable successor.

Key Points

• Meta and Shanghai based Fourier reportedly built Max in about three months.
• The prototype stands about 1.5 metres tall, weighs 48 kilograms and uses Nvidia Jetson Thor computing hardware.
• Max has a facial display, chest speakers and hands that reportedly grasp small bottles and toys.
• Meta's interaction targets include recognising speech at five metres and gestures at three metres.
• Max 2 entered development in September, with improved hands, a redesigned torso, a projector and sensors for stairs planned.
• Meta says the collaboration is for internal research and experimentation, with no plans to sell these units.