0

TPL_YOUR_BAG

Cart empty

A human entering a workplace can instantly interpret dozens of things without consciously thinking about them.

A door is closed.

A box is on the floor.

A colleague is approaching.

A tool has been left on a table.

A particular object needs to be moved.

The environment is not perfectly organized.

Yet a person can adapt.

This is the challenge that MATRIX-1 by Matrix Robotics is designed to address.

Rather than focusing exclusively on movement, MATRIX-1 combines advanced AI reasoning, visual perception, tactile manipulation and autonomous navigation into a full-size humanoid platform.

The manufacturer describes MATRIX-1 as a next-generation AI humanoid designed for safe and efficient real-world tasks, with autonomous capabilities intended for complex environments.

China Robot Store currently lists MATRIX-1 at $25,200, with a height of 180 cm, weight of 67 kg, 10 kg payload per arm, up to 5 hours endurance, and a listed maximum speed of 7.6 km/h.

The Next Problem After Walking

The first generation of humanoid demonstrations concentrated heavily on locomotion.

Can the robot:

Stand?

Walk?

Turn?

Recover balance?

These remain important engineering problems.

But once a robot can move around, a much harder problem emerges.

What should it do?

A useful humanoid must understand its environment well enough to decide:

  • Where to go
  • What object to pick up
  • How to grasp it
  • Which route to take
  • How to interact with people
  • When to stop
  • When human assistance is necessary

This is where MATRIX-1 attempts to move beyond traditional robotics.

Four Times More AI Reasoning

One of the most distinctive claims surrounding MATRIX-1 is its reasoning AI module.

Matrix Robotics states that onboard computing and AI reasoning capability have increased by 4× compared with the previous generation.

The important concept is not simply raw computing power.

It is what the robot can do with that computing.

The manufacturer positions MATRIX-1 as capable of autonomous execution of real-world AI tasks.

In other words:

See → Understand → Decide → Act

rather than simply:

Receive command → Repeat programmed movement

Why Reasoning Matters in Physical AI

Digital AI can generate a response to a question.

A physical robot has to do something about the answer.

Imagine a person telling the robot:

“Move this package to the other side of the room.”

The instruction sounds simple.

But the robot must solve a chain of physical problems.

First:

What is “this package”?

Then:

Where is the other side of the room?

Then:

Can I reach it safely?

Then:

How should I grasp it?

Then:

Which route should I use?

Then:

Where exactly should I place it?

This is what makes Physical AI fundamentally different from ordinary software AI.

The answer has consequences in the real world.

Vision-Language Models Enter the Robot

MATRIX-1 incorporates an onboard Vision-Language Model (VLM) capable of visual reasoning using the robot's cameras.

This is an important development in humanoid robotics.

A conventional vision system might classify:

“Box.”

A VLM-based robotic system aims to interpret something closer to:

“There is a cardboard box beside the table, and it appears to be the object relevant to the current instruction.”

That difference is enormous.

The robot moves from recognizing isolated objects toward interpreting the meaning of objects in context.

Eight Cameras and a Physical World

MATRIX-1 uses eight onboard RGB cameras for AI-driven visual perception. The system can also incorporate LiDAR and ultrasonic sensors for obstacle detection.

This creates a multi-layer perception system.

The robot needs to understand:

Objects

People

Distances

Obstacles

Movement

Free space

Physical relationships

The result is an increasingly rich representation of the environment.

Seeing Is Not Enough

There is another important aspect.

A robot can see a cup.

But seeing a cup does not tell it exactly how much force should be used to pick it up.

This is where MATRIX-1's dexterous hands become particularly important.

The robot uses third-generation five-fingered hands with 22 degrees of freedom, tactile sensing and force-position hybrid control.

The objective is to give the robot a more human-like ability to manipulate objects.

22 Degrees of Freedom in the Hands

The human hand is extraordinarily difficult to reproduce mechanically.

It can:

  • Rotate
  • Bend
  • Press
  • Grip
  • Pinch
  • Slide
  • Stabilize
  • Adjust force continuously

MATRIX-1's 22-DOF dexterous hand is designed to provide a high level of articulation for robotic manipulation.

The significance is not the number alone.

The real value comes from combining:

22 DOF


Tactile sensing


Force-position control


AI perception

Together, these technologies create the possibility of more sophisticated physical interaction.

Force-Position Hybrid Control

Robots traditionally operate using position control.

The system tells the robot:

“Move your hand here.”

But physical interaction often requires something more.

Imagine inserting a connector.

The robot needs to know not only where the connector is, but also whether it is encountering resistance.

Force-position hybrid control combines these two dimensions.

The robot can reason about:

Where should I move?

and

How much force should I apply?

This is essential for delicate manipulation.

Tactile Sensing Gives the Robot a Sense of Touch

Vision provides information before contact.

Tactile sensing becomes valuable during contact.

Imagine MATRIX-1 picking up an unfamiliar object.

The cameras estimate:

Shape + Size + Position

The fingers then provide additional information:

Contact + Pressure + Interaction

The robot can combine both.

That makes tactile sensing a key part of the transition from visual AI to physical intelligence.

10 Kilograms Per Arm

MATRIX-1 is listed with a payload of approximately 10 kg per arm.

This gives the robot enough capacity for a broad range of practical manipulation tasks.

Potential applications include:

  • Material handling
  • Packaging
  • Component transfer
  • Inspection
  • Tool manipulation
  • Object sorting
  • Logistics assistance
  • Service operations

But the real objective is not simply to lift 10 kg.

It is to lift the correct object in the correct way.

Human-Like Flexibility

Matrix Robotics has developed its own motors and actuators for MATRIX-1.

The manufacturer highlights a wide range of joint motion intended to produce highly flexible and human-like movement.

This matters because real environments are rarely perfectly optimized for robots.

A humanoid might need to:

  • Bend around an object
  • Rotate its torso
  • Reach sideways
  • Change its posture
  • Work at different heights
  • Move through tight spaces

Flexibility becomes a functional advantage.

A Soft Exterior for Human Interaction

MATRIX-1 also incorporates a new motion-enclosure mechanism using specialized enclosure materials around the neck and torso.

Matrix Robotics states that the design is intended to improve flexibility while also contributing to interaction safety.

This is particularly relevant to collaborative robotics.

The future humanoid will not always work behind a safety cage.

It may operate directly beside humans.

A Robot That Can Open Doors

One of the manufacturer's more interesting examples is surprisingly ordinary:

opening doors.

MATRIX-1 is designed to autonomously handle different types of door handles and navigate through unstructured spaces.

This sounds trivial.

It is not.

Doors are a perfect example of why general-purpose robotics is difficult.

Different buildings have:

  • Lever handles
  • Round knobs
  • Push bars
  • Heavy doors
  • Light doors
  • Automatic doors
  • Partially obstructed entrances

A robot must identify the mechanism and adapt its manipulation strategy.

That is a general-purpose robotics problem in miniature.

Unstructured Environments Are the Real Test

Factories are becoming increasingly automated.

But many real-world environments remain unpredictable.

An office can change every day.

A warehouse can contain new objects.

A public building can have unfamiliar layouts.

A service environment can contain unpredictable human behavior.

A general-purpose humanoid must therefore operate beyond perfectly controlled conditions.

MATRIX-1 is specifically designed around this concept of navigating unstructured environments.

Autonomous by Default

Matrix Robotics states that MATRIX-1 operates autonomously by default, using AI to explore and operate within a workspace.

The robot can recognize people and objects from a distance and perform physical tasks while monitoring the environment.

This creates a different model of robotics.

Instead of:

Human controls every movement

the objective becomes:

Human gives objective

Robot interprets environment

Robot plans

Robot acts

Human supervises when needed

That is much closer to the concept of an autonomous robotic worker.

Shared Autonomy

Complete autonomy will not always be appropriate.

There will be situations where a robot encounters something unexpected.

Instead of failing completely, MATRIX-1 can use a shared-autonomy approach in which a human operator can intervene when necessary.

This could become an important bridge between today's robotics and fully autonomous systems.

The robot handles routine situations.

The human handles exceptional ones.

Why This Model Could Work

Consider a robot working in a large office building.

Most tasks are repetitive:

Patrol

Inspect

Move objects

Check areas

Report anomalies

But occasionally something unusual happens.

A door is damaged.

An object blocks a passage.

An unfamiliar situation appears.

Instead of stopping permanently, the robot requests assistance.

A human operator takes control.

The robot continues afterward.

This creates a practical compromise between autonomy and human supervision.

The 5-Hour Endurance Question

MATRIX-1 is currently specified for approximately 5 hours of operating time, with a 2.28 kWh battery pack and autonomous charging-base support.

This is shorter than some industrial humanoids designed around full-shift operation.

But endurance is not an isolated specification.

A useful commercial robot also needs to consider:

Task intensity

Charging time

Autonomous charging

Downtime

Robot utilization

A robot that can autonomously return to a charging station can potentially remain useful without requiring constant human intervention.

Fast Charging

Matrix Robotics states that MATRIX-1 can charge for approximately 45 minutes to operate for up to 5 hours.

If achieved under appropriate operating conditions, that ratio could be particularly useful for commercial deployments.

The key concept is:

Short charging window → Long operating period

This can increase the amount of time the robot is actually available for work.

7.6 km/h Changes Mobility

MATRIX-1's manufacturer lists a maximum speed of 7.6 km/h, while its main English product page presents a nominal speed of 1.2 m/s.

This discrepancy illustrates why robot specifications should always be interpreted carefully: maximum speed, nominal speed and safe operating speed are not necessarily the same thing.

Still, the platform is designed for meaningful mobility rather than simply standing at a workstation.

General-Purpose Robotics

The manufacturer's positioning is particularly important.

MATRIX-1 is not designed exclusively as:

Factory robot

or

Home assistant

or

Security robot

It is presented as a general-purpose humanoid robot.

This means the same platform is intended to support multiple categories of physical work.

That is exactly where the humanoid concept becomes economically interesting.

One Platform, Multiple Industries

Potential applications include:

Manufacturing

Material handling, assembly support and inspection.

Logistics

Object movement, sorting and warehouse assistance.

Service

Guidance, facility support and customer-facing interaction.

Security

Patrol and anomaly detection.

Research

Embodied AI and robotic learning.

Human Assistance

Physical tasks that require manipulation and mobility.

The objective is to make the robot adaptable rather than highly specialized.

The Importance of Learning

MATRIX-1 uses imitation learning and reinforcement learning as part of its AI approach.

This matters because general-purpose robots cannot be manually programmed for every possible situation.

A robot may need to learn:

How to grasp

How to walk around obstacles

How to interact with unfamiliar objects

How to respond to changing conditions

Learning-based robotics provides a path toward this flexibility.

From Demonstration to Skill

There is a major difference between:

“The robot performed this task once.”

and:

“The robot has learned the skill.”

The first is a demonstration.

The second is a capability.

For Physical AI to become commercially useful, robots need to accumulate reusable skills.

One learned behavior should ideally become a building block for many related tasks.

Why the Hands Could Be More Important Than the Legs

The public image of humanoid robotics often centers around walking.

But for many commercial applications, the hands could ultimately determine productivity.

A robot can walk through a warehouse all day.

If it cannot reliably manipulate objects, its usefulness remains limited.

MATRIX-1's combination of:

22-DOF hands

Tactile sensing

Force-position control

VLM perception

is therefore one of its most important technological features.

A New Definition of “Robot Intelligence”

Traditional robots are often measured by:

Accuracy

Speed

Payload

Repeatability

AI robots add new measurements:

Understanding

Adaptability

Reasoning

Generalization

Learning

The future robot will need both.

A machine that thinks well but moves badly is not useful.

A machine that moves perfectly but cannot understand changing circumstances is also limited.

The goal is:

Intelligence + Physical Capability

MATRIX-1 and the Physical AI Era

This is where MATRIX-1 becomes part of a much larger industry trend.

The AI industry is increasingly moving from models that operate only on screens toward systems that operate in physical environments.

The robot becomes the body.

The AI becomes the reasoning layer.

The sensors become the perception system.

The actuators become the physical interface.

Together they form a new category:

Physical AI.

The Future Robot May Not Follow Scripts

Imagine a robot receiving an instruction:

“Check the conference room and prepare it for the next meeting.”

A traditional robot would struggle with the ambiguity.

A general-purpose AI robot could potentially interpret the instruction as a collection of tasks:

  • Navigate to the room
  • Check whether people are present
  • Identify objects that are out of place
  • Move them
  • Check the table
  • Report anything unusual

This is the kind of open-ended task execution that embodied AI is trying to make possible.

The Human Is Still in the Loop

Despite the advances, autonomous humanoid robots are not magic.

Real-world environments remain extremely complicated.

Unexpected objects.

Unusual lighting.

Human behavior.

Mechanical failures.

Unfamiliar tasks.

Safety requirements.

The most realistic future is therefore likely to combine:

Autonomy + Human Oversight

rather than eliminating humans completely.

MATRIX-1's shared-autonomy concept fits this model.

MATRIX-1 at China Robot Store

MATRIX-1 is currently listed in the China Robot Store Humanoid Robots catalog.

Current listed price: $25,200

Height: 180 cm

Weight: 67 kg

Payload: 10 kg per arm

Hands: 22-DOF five-finger dexterous hands

Speed: up to 7.6 km/h according to manufacturer specifications

Endurance: up to 5 hours

Cameras: 8 RGB cameras

AI: VLM + AI reasoning

Navigation: autonomous perception and obstacle avoidance

Charging: autonomous charging base

Application: general-purpose humanoid robotics

Explore MATRIX-1:

https://china-robot.store/humanoid/matrix-1-detail

The Real Question Is No Longer “Can It Walk?”

The humanoid robotics industry has passed that question.

The more important questions are now:

Can it understand?

Can it manipulate?

Can it adapt?

Can it learn?

Can it recover from unexpected situations?

Can it safely operate around people?

MATRIX-1 is built around these questions.

From Machine to Physical AI Agent

The long-term evolution of robotics may look something like this:

Robot 1.0

Pre-programmed machine.

Robot 2.0

Sensor-driven autonomous machine.

Robot 3.0

AI-powered adaptive robot.

Robot 4.0

General-purpose Physical AI agent.

MATRIX-1 is part of the movement toward the third and fourth stages.

Its goal is not simply to repeat movements.

It is to understand enough of the physical environment to make decisions and execute tasks.

What Makes MATRIX-1 Different

The strongest feature of MATRIX-1 is not one specification.

It is the integration of several technologies:

4× AI reasoning


Vision-Language Model


8-camera perception


22-DOF tactile hands


Force-position control


Autonomous navigation


Human-like mobility

This is the architecture required if humanoid robots are going to become useful beyond controlled demonstrations.

The Next Step: Real Deployment

The real test for MATRIX-1 will not be a stage demonstration.

It will be sustained operation in environments that are:

Messy

Dynamic

Unpredictable

Human

That is where the difference between a prototype and a useful robot becomes visible.

And that is also where the future of Physical AI will be decided.

MATRIX-1 and the Future of Humanoid Robotics

The next generation of humanoid robots will increasingly compete on intelligence rather than appearance.

Walking is becoming expected.

Manipulation is becoming expected.

Vision is becoming expected.

The new frontier is:

Reasoning in the physical world.

MATRIX-1 represents one approach to this challenge.

A full-size humanoid robot with AI reasoning, tactile hands, multimodal perception and autonomous mobility—designed to operate in the same environments where humans already work.

That is the real significance of MATRIX-1.

Explore Advanced Humanoid Robots

Discover the complete Humanoid Robots collection:

https://china-robot.store/humanoid

Explore AI-powered robotics from China and around the world:

https://china-robot.store/

China Robot Store News

China Robot Store News follows humanoid robotics, Physical AI, embodied intelligence, AI reasoning, dexterous manipulation and next-generation automation.

Our NEWS series focuses on real robotic platforms and the technologies moving AI from digital systems into the physical world.