The latest generation of humanoid, industrial and service robots shows how quickly physical AI is advancing.
BEIJING — The robot revolution is beginning to look a lot less like science fiction and a lot more like a workforce.
The 2026 World Robot Conference in Beijing wrapped up this week after five days of demonstrations that offered a striking glimpse at where robotics may be heading next.
Humanoid robots boxed. Others danced and played table tennis. Robots packed smartphones, sorted packages and stocked shelves. One company demonstrated a human-sized robot designed specifically for companionship and elder care.
And then there was the robot horse — a massive four-legged machine capable of carrying a human rider.
The conference ran from August 19 through August 23 and featured roughly 3,000 robotic products. Organizers say 44 new technologies and solutions were unveiled, while more than 300 exhibitors participated in the broader showcase.
The takeaway was difficult to miss:
Robotics is rapidly moving from machines that demonstrate what is technically possible to machines being designed to perform actual jobs.
Qiji X1: A Robot Horse You Can Actually Ride
One of the strangest — and perhaps most futuristic — machines attracting attention was the Qiji X1 from DaxAI Robotics.
Instead of wheels, the enormous machine moves using four articulated mechanical legs. A human rider sits on top using handlebars, essentially turning the robot into a mechanical horse.
DaxAI demonstrated the Qiji X1 carrying a person across the conference floor.
The machine is reportedly capable of supporting approximately 300 kilograms, or about 660 pounds, potentially opening the door to transportation across terrain where conventional wheeled vehicles struggle.
The concept could eventually have applications in agriculture, emergency response, industrial sites, military logistics and rugged transportation.
For now, however, watching someone ride a giant walking robot remains one of the more surreal images of the conference.
Unitree Robots Are Fighting, Dancing and Playing Sports
Unitree continued demonstrating just how athletic humanoid robots are becoming.
At its conference booth, humanoid robots boxed one another, performed synchronized dances and even played table tennis against a human opponent.
The demonstrations are entertaining, but the underlying technology is important.
Boxing and athletic movements require robots to constantly calculate balance, body position, impact forces and recovery. A robot that gets bumped or misses a movement must quickly stabilize itself instead of simply falling over.
Those capabilities could eventually translate into robots capable of safely navigating unpredictable workplaces.
The demonstrations also highlight how quickly humanoid hardware has advanced. Robots that struggled simply to walk several years ago are increasingly capable of running, jumping, kicking and recovering from complex movements.
Robotera Is Already Sorting Packages
Perhaps more significant than robot boxing was what Robotera demonstrated.
Its robotics systems are being developed for logistics operations, including sorting packages traveling through warehouses.
According to company representatives cited in reporting from the conference, more than 100 Robotera parcel-sorting robots are operating across 15 warehouses, including logistics facilities associated with China Post.
That distinction matters.
These aren’t simply experimental robots performing a carefully scripted trade-show demonstration.
They represent the beginning of robots performing repetitive physical jobs inside real commercial environments.
Warehouse automation has traditionally relied heavily on conveyor belts, robotic arms and autonomous carts. Humanoid or semi-humanoid robots could eventually provide something those systems cannot easily deliver: machines capable of working inside environments originally designed around human workers.
DexForce Robots Are Packing Smartphones
Another demonstration provided an even clearer picture of what robotic manufacturing could look like.
DexForce showed its wheeled humanoid robots packing smartphones into boxes.
The company’s W1 Pro robots are reportedly already operating on a production line at Lens Technology, a major electronics supplier.
The system can identify when a phone is positioned incorrectly and adjust it before packaging, with the company claiming millimeter-level precision.
It sounds like a relatively simple job.
That’s exactly why it matters.
The first major wave of humanoid automation may not involve robots performing spectacular tasks. It could involve thousands of machines taking over repetitive activities such as picking, sorting, packing, loading and inspecting products.
Galbot G1 Wants to Work in Stores and Warehouses
Galbot showed another approach to physical AI.
Its G1 uses wheels rather than humanoid legs, while maintaining a torso and two robotic arms capable of interacting with objects.
Galbot is targeting logistics and service environments where robots could retrieve products, move goods and handle inventory.
Reuters highlighted Galbot among the companies demonstrating practical robotics applications at this year’s conference.
The wheeled design also raises an interesting question for the robotics industry:
Do robots actually need legs?
For factories, hospitals, pharmacies and retail stores with relatively flat floors, wheels may provide a cheaper, faster and more energy-efficient solution.
The future robot workforce may therefore include everything from humanoids to wheeled assistants and specialized robotic arms.
Fourier GR-3: A Robot Designed to Care for People
Not every robot at the conference was designed for factories.
Fourier’s GR-3 takes almost the opposite approach.
The approximately human-sized robot has been designed as what Fourier calls a “Care-bot.”
Rather than exposing metal and mechanical components, GR-3 uses softer materials intended to make interactions with humans feel less intimidating.
The robot incorporates vision, microphones and touch sensors to recognize people and respond to interactions. Fourier says potential applications include companionship, education, public services, rehabilitation and eventually elder care.
GR-3 stands approximately 165 centimeters tall, weighs around 71 kilograms and offers up to 55 degrees of freedom.
The idea points toward another potentially enormous robotics market: caring for aging populations.
If humanoid robots eventually become capable of reliably helping older adults with basic tasks, mobility and companionship, the economic and social impact could be significant.
LimX Luna Can Learn Movements From Video
LimX Dynamics brought one of the conference’s most visually impressive humanoids.
The LimX Luna stands about 160 centimeters tall and has 27 degrees of freedom.
It can run, jump, perform complex movements and learn movements from video. LimX also built Luna around multimodal interaction, allowing the robot to respond to speech, visual information and gestures.
But one feature stands out.
LimX says its control system can coordinate more than 200 Luna robots simultaneously, synchronizing their movements with millisecond-level precision.
That opens possibilities for entertainment, theme parks, live events, hospitality and interactive commercial environments.
Instead of programming every movement manually, future robots could increasingly learn actions from humans or video demonstrations.
That could dramatically reduce the time required to teach robots new tasks.
SIASUN Shows Robots Still Have a Heavy-Industry Future
While humanoids received most of the attention, traditional industrial robotics continues advancing as well.
SIASUN displayed its massive SR270A industrial robotic arm.
The machine can handle a 270-kilogram payload with a reach of approximately 2.7 meters.
It is designed for jobs including handling, assembly, loading, unloading and spot welding.
Robotic arms like these aren’t as visually exciting as humanoids performing backflips, but they remain the backbone of industrial automation.
And they demonstrate an important point.
The future of robotics probably won’t belong to one universal robot.
Instead, specialized machines will increasingly work alongside humanoids, autonomous vehicles, robotic arms and AI-powered software systems.
The Bigger Story Isn’t the Backflips
Watching robots fight or perform synchronized dances makes great video.
But those demonstrations aren’t necessarily the biggest robotics story coming out of Beijing.
The bigger story is the mundane stuff.
Sorting packages.
Packing phones.
Restocking shelves.
Moving factory components.
Those are tasks businesses currently pay millions of people around the world to perform.
And robots are beginning to do them.
China has made robotics a strategic industrial priority, and more than 150 companies are now involved in the country’s humanoid robotics sector. However, significant limitations remain. Current humanoids can still struggle with dexterity, adapting to unfamiliar environments and performing general-purpose work without extensive training or human assistance.
Even Unitree CEO Wang Xingxing has cautioned that the breakthrough where humanoids can reliably complete most unfamiliar physical tasks from simple language instructions may still be years away.
That is an important reality check.
These machines are impressive.
They are not yet replacements for humans across most jobs.
But Something Has Clearly Changed
For decades, the promise of general-purpose robots has always seemed to exist somewhere in the future.
This year’s World Robot Conference suggested that future is beginning to move closer.
The question is no longer simply:
“Can we build a humanoid robot?”
We clearly can.
The next questions are much bigger:
Can they become reliable?
Can companies manufacture them cheaply?
Can artificial intelligence teach them thousands of tasks instead of one carefully programmed routine?
And perhaps most importantly:
What happens to the workforce when the answer to all three becomes yes?
Robots aren’t just learning how to walk anymore.
They’re learning how to work.
And that could make the next decade of AI look very different from the last one.