America's top humanoid robots are locked in a swagger battle.
In Silicon Valley, the $39 billion startup Figure AI has been livestreaming a small fleet of its AI-powered humanoid robots sorting packages in a mock logistics center.
The AI robots are powered by the startup's self-developed Helix 02 neural network that runs onboard. When a robot's battery runs low after about 4 hours, it heads to a wireless charging station as a humanoid robot colleague takes over.
It started as an 8-hour marketing stunt but Figure let it run nonstop as the robots kept going. To liven things up, the robotics firm pitted an intern against the robot in a 10-hour package sorting contest. The human intern sorted nearly 13,000 packages, beating the Figure 03 robot by just under 200 items. Figure founder Brett Adcock predicted that it's the last time a human will win the challenge.
Meanwhile in Oregon, Agility Robotics pointed out that it livestreamed its Digit humanoids doing autonomous warehouse work years ago and that they're scaling their deployment at actual worksites. Like Figure, Agility has a dedicated robot factory at its headquarters. Figure's San Jose plant has a reported annual capacity of 12,000 units, while Agility's is 10,000.
On the East Coast, Boston Dynamics entered the chat, sharing footage of its world-famous Atlas humanoid robot smoothly lifting a loaded minifridge. The Massachusetts robotics leader said Atlas used its full-body coordination and dynamic balance to neatly place the unit on a table next to a distracted employee. The Hyundai subsidiary called it proof that Atlas is the world's most capable humanoid robot for industry.
And as Tesla prepares to unveil its highly anticipated Generation 3 Optimus humanoid robot, a Chinese startup launched by one of its former engineers is already beating it to mass production.
0:00 - America’s Humanoid Robot Race Heats Up 0:54 - Figure 03 Humanoids Livestream Warehouse Work 1:36 - Are Figure’s Robots Really Autonomous? 1:51 - Figure AI's I, Robot (2004) Inspired Vision 3:43 - How Figure’s Helix 02 VLA AI Model Works 4:26 - Human vs. Robot Package Sorting Contest 5:28 - China’s Robot Era Enters Logistics Centers 5:53 - Agility Robotics and Digit Respond 6:11 - Boston Dynamics' Atlas Lifts a Mini Fridge 6:40 - Atlas Uses Simulation and Digital Twins 7:03 - Atlas Design, Damage, and Industrial Specs 8:04 - Unitree’s Mass Production Challenge 9:02 - Hyundai Pushes for Atlas Factory Deployment 10:20 - Where Is Tesla Optimus? 10:43 - China's Matrix Robotics Unveils Matrix-3 12:33 - China’s Humanoid Startups Gain Global Attention
The global race for general-purpose humanoid robots is intensifying, with Silicon Valley startups, established industry leaders, and Chinese firms competing for dominance. Companies like Figure, Boston Dynamics, and Matrix Robotics are leveraging advanced AI and simulation to move robots from controlled labs into real-world logistics and manufacturing roles.
America's top humanoids are locked in a swagger battle
as the race toward general-purpose robotics heats up.
Silicon Valley's top-funded robotics startup is evoking iRobot,
the movie, not the bankrupt robot vacuum maker,
as it insists its autonomous humanoids are ready for real-world labor.
Meanwhile, the oldest company in the humanoid race
is going to great lengths to show it's agile enough
to compete with its much younger rivals.
And as Tesla prepares to unveil its highly-anticipated
Generation 3 Optimus humanoid,
a startup launched by one of its former engineers
is already beating it to mass production.
In Silicon Valley, the $39 billion startup figure
has been live-streaming a small fleet of its humanoids,
working around the clock in a mock logistics facility.
A week into the stream,
the robots had sorted nearly 200,000 parcels.
They took turns working the conveyor-style setup
at Figures headquarters in San Jose.
They handled a range of package types
to make sure their barcodes were facing down
on the looping conveyor belt.
When one robot worker's battery ran low after about four hours,
it went to a wireless charger while another took over.
The stream attracted millions of viewers across platforms.
While many cheered the robots on,
others suspected they were teleoperated
because they spotted some odd human-like pauses and gestures.
In an interview with Bloomberg,
figure founder Brett Adcock insisted
they were fully autonomous.
The billionaire founder also referenced the 2004 movie
iRobot to explain his vision.
The movie is loosely inspired by science fiction author
Isaac Asimov's book of short stories of the same name.
It's set in 2035 Chicago,
where humanoid robots are everyday helpers
across sectors and in households.
The androids, developed by a company called U.S. Robotics,
operate under a strict set of rules
called the Three Laws of Robotics,
which requires them to avoid harming humans at all costs.
Will Smith plays an anti-clanker detective
who suspects a humanoid robot named Sunny
is responsible for a homicide.
His investigation uncovers that the company's
central artificial intelligence system
has interpreted the Three Laws in a dangerous way.
To protect humanity from itself,
it decides that robots must take control of society.
The technology in the Alex Proyas film
is far more advanced than what's possible today,
but parts of it are starting to look more realistic.
Minus the dystopian outcome,
the world of iRobot matches what figure's founder describes
as a future of abundant labor.
The Figure 3 is an early real-world step toward that idea,
but much more limited.
It can perceive its surroundings,
use AI to plan movements,
and perform basic tasks within controlled settings,
but with far less flexibility than the robots in the movie.
The future presented in iRobot
feels unlikely to materialize by 2035,
but breakthroughs are rapidly moving robots
toward the vision faster than many expected.
Snapping back to reality,
Figure's robots are powered by the startup's
Helix two-vision language action model.
The AI-controlled model acts
as the Figure 3's brain and nervous system.
The robot's onboard artificial intelligence
generates hardware commands
using sensor data on vision, touch, and body position.
Figure is aggressively collecting data to feed the model,
as its founder believes that a massive amount of real-world data
is needed to unlock general autonomy.
Their ultimate goal is for the robot
to enter any unfamiliar scenario
and adapt just like a person.
The stream started as an eight-hour marketing stunt,
but Figure kept it going as the robots continued working.
To spice things up,
Figure pitted an intern against the robots
in a 10-hour package sorting contest.
The college student narrowly defeated the Figure 3 robots
by just under 200 packages.
He sorted nearly 13,000 parcels
at an average speed of 2.79 seconds per item,
while the robots averaged at 2.83 seconds.
Launched in 2022,
Figure says it's developing its humanoids
for use across sectors and eventually in households.
They're targeting a price of around $20,000 per unit
after production scales.
The startup is focusing heavily on logistics work,
as it's reportedly in talks to supply UPS
with a large number of humanoids.
They're not alone.
In China,
Beijing's Robot Era has reportedly deployed
its flagship humanoids in more than 10 logistics centers
at major delivery operators like SF Express and China Post.
The startup, established in 2023,
reports that the autonomous machines
have already reached around 85% human-level efficiency
in some facilities while running continuously.
Meanwhile in Oregon,
Agility Robotics pointed out
that it live-streamed its Digit humanoids
doing autonomous warehouse work years ago,
and that they're scaling their deployment
at actual work sites.
On the East Coast,
Boston Dynamics entered the chat,
sharing this footage of its world-famous Atlas humanoid
smoothly lifting a loaded mini-fridge.
The Massachusetts robotics leader said
Atlas used its full body coordination and dynamic balance
to neatly place the unit on a table
next to a worker pretending to be distracted.
On social media,
the company called its latest demo,
which also included some superhuman acrobatics,
proof that Atlas is the most capable humanoid robot
for industry.
According to Boston Dynamics,
Atlas was trained first in a simulation.
Where its digital twins finished
millions of hours of virtual practice
to find the best way to lift and move the fridge
that weighs about 50 kilos or 110 pounds
without dropping it or losing balance.
On a negative note,
Atlas' grippers dinged up the appliance
just from picking it up.
The demo also highlighted
the humanoid's unusual body design,
which has 56 degrees of freedom
and can rotate way beyond normal human limits.
It also has a dual hot swappable battery system,
similar to industrial humanoids
like UB Tech's Walker S2 and Hexagon's Eon.
The Hyundai subsidiary is scaling production of Atlas,
which began as a DARPA project in 2013.
Mark Rayburt launched Boston Dynamics
as an MIT spinout in 1992
after pioneering research into legged robots
at the Institute's Leg Laboratory.
Despite its massive cultural impact,
the company has historically sold
a relatively small number of robots,
primarily Spot, its $75,000 quadruped
and its logistics system Stretch.
As Boston Dynamics commercialized Spot in 2019,
the Chinese company Unitree
aggressively moved toward mass production.
Launched in 2016 by hardware engineer Wang Xingxing,
Hangzhou-based Unitree reported
shipping more than 18,000 robodogs
and 5,500 humanoids in 2025 alone.
Unitree's founder, shown here
climbing into his company's $650,000 mecha robot,
has cited the legendary robotics firm
as one of his biggest influences.
But Boston Dynamics sees them
as a potential national security threat.
The company recently sent one of its VPs
to testify for Congress
about its biggest overseas rival.
As it enters its mid-30s,
Boston Dynamics faces immense pressure
to prove it can keep up with its younger competitors.
Its parent company Hyundai reportedly
wants tens of thousands of Atlas humanoids
deployed at the South Korean automaker's factories
as soon as possible.
Boston Dynamics has also committed
to delivering humanoids to the Hyundai subsidiary Kia,
as well as its ex-parent company Google,
through its partnership with DeepMind.
According to a recent report from Semaphore,
the company's board of directors
has pushed executives out,
as it's grown critical of its narrowing lead
in the humanoid race.
Semaphore reported that Boston Dynamics
has been building about four
of its Atlas humanoids per month
as it prepares to open a new robot factory.
The robotics firm appeared quick
to respond to Semaphore's reporting,
with this demo of the latest version of Atlas
intended for mass production.
The company first introduced
the new and improved humanoid
as a stationary display at CES 2026.
In its first public-facing demo,
the robot performed a series
of highly dynamic gymnastic
and balance-focused maneuvers.
Boston Dynamics captioned the footage saying,
balancing commercial goals and robotics can be tricky,
but they're making it work with Atlas.
The flurry of activity in the U.S. humanoid sector
has many asking,
where's Tesla?
As the EV maker prepares to unveil
its highly anticipated
third-generation Optimus humanoid,
a startup launched
by one of its former engineers in China
is already racing toward mass production.
Alan Zhang, who reportedly played a major role
in the Tesla Bot program,
founded Matrix Robotics in Shanghai in 2024.
The two-year-old robotics firm
just unveiled what it calls
its mass-production-ready iteration
of its flagship humanoid,
the Matrix 3.
Standing 170 centimeters,
or 5 feet 7 inches,
the biped robot weighs around 65 kilos,
or 143 pounds.
It's covered in 3D-woven biomimetic skin
with embedded tactile sensors
that's sensitive to pressures
as light as a gentle fingertip touch.
The fake human's five-fingered hands
have a reported 27 degrees of freedom,
which is in the upper echelon
of dexterous end-effectors
hitting the market today.
The Matrix hand has a tendon-driven design
that's inspired by human anatomy.
On paper, its dexterity
is comparable to a human's hand,
which also typically has
around 27 degrees of freedom.
The Matrix 3 runs on the startup's
in-house physical model called Wave.
According to the robotics firm,
the system enables zero-shot generalization
so the humanoid can understand
basic physical principles,
interpret natural language instructions,
plan actions,
and adapt to novel tasks
in unstructured environments.
The artificial intelligence combines
input from the robot's cameras
with thousands of tiny sensors
across its electronic skin and fingertips.
The Matrix plans to start
large-scale manufacturing
in the second half of 2026
at its new Shanghai factory,
which has a planned annual capacity
of 10,000 units.
The Matrix 3's pricing
starts around $100,000.
Meanwhile in China,
the country's top-funded humanoid startup
has reached a new level of visibility
outside the mainland
after a viral encounter with Fox News.
For more on the future
as envisioned by Beijing's Galbit,
check the related video.