X Square Robot is showcasing its full portfolio of embodied AI applied sciences on the World Robot Conference 2026 in Beijing, presenting purposes starting from logistics automation and dexterous manipulation to family robotics and AI training-data assortment.
The Shenzhen-based firm is exhibiting at sales space C107 throughout WRC, which runs from August 19 to 23, bringing collectively its WALL-B embodied AI basis mannequin, robotic {hardware}, manipulation applied sciences and knowledge infrastructure.
The WRC look comes shortly after X Sq. Robotic livestreamed an hour-long logistics demonstration by which its system processed 1,816 parcels with a reported accuracy of greater than 98 %.
At WRC, the corporate is increasing the main focus past that logistics utility to display how the identical broader strategy to embodied intelligence will be utilized throughout totally different bodily methods and environments.
Embodied AI throughout totally different purposes
One of many central demonstrations on the X Sq. Robotic sales space is the logistics system featured within the current livestream.
The system combines the corporate’s WALL-B basis mannequin with its self-developed high-performance six-axis robotic arms to deal with parcels of various sizes, weights, shapes and supplies.
However the firm’s WRC presentation extends from industrial operations into rather more advanced manipulation duties.
In a flower-arranging demonstration, a dual-arm robotic is given natural-language directions that require it to grasp a buyer’s desire for a specific colour of rose, establish the suitable flower after which full a sequence involving a vase, extra flowers and foliage.
The robotic should coordinate notion, language understanding and two-arm manipulation throughout the entire process.
Modifications can be launched into the setting, together with transferring the vase or rearranging the flowers, testing whether or not the system can adapt its actions relatively than merely replay a predetermined sequence of actions.
Fan disassembly with dexterous fingers
The fan-handling demo is designed to check how a robotic can use its embodied AI basis mannequin and dexterous hand to finish an extended sequence of exact manipulation duties.
The robotic must open a cardboard field, retrieve a fan, place it on the desk, activate the change, and confirm that it’s working. Alongside the way in which, it should deal with deformable packaging, restricted visibility, and an irregularly formed object, whereas coordinating each arms and exactly controlling its five-finger dexterous hand.
Behind the demo is X Sq. Robotic’s general-purpose talent technology platform for dexterous fingers, which integrates knowledge administration, talent coaching, and real-robot analysis.
It helps expertise equivalent to greedy, turning, opening and shutting, and power use, connecting the workflow from demonstration and knowledge processing to mannequin coaching, deployment, and analysis.
By turning uncooked demonstrations into learnable and constantly improvable talent property, the platform helps robots transfer past particular person duties towards constructing transferable, general-purpose manipulation capabilities.
Bringing embodied AI into the house
X Sq. Robotic can also be utilizing WRC to display its “X Household Member Program” by a real-life residence setting constructed across the idea of “a day at residence”.
Somewhat than presenting particular person robotic features in isolation, the demonstration connects totally different on a regular basis eventualities, together with leisure, household meals, leaving the house and remotely interacting with the house by an app.
The idea is meant to display how robotics, AI, linked gadgets and providers may work collectively as individuals transfer between totally different actions throughout the day.
It additionally presents a considerably totally different problem for embodied AI from the comparatively managed circumstances of an industrial workstation.
Properties include altering preparations of objects, individuals and actions, making notion and flexibility significantly essential if robots are ultimately to function autonomously in home environments.
Constructing the information behind embodied AI
Underlying these purposes is one other main a part of X Sq. Robotic’s WRC portfolio: the QUANXTA Zero collection, a hardware-and-software platform for producing embodied AI coaching knowledge.
The product household consists of totally different combos of head-mounted gear, wearable {hardware} and handheld grippers designed to seize human actions with out requiring the operator to be bodily linked to a robotic.
In response to X Sq. Robotic, the system synchronizes a number of sensor streams to inside 1 millisecond and may seize imaginative and prescient, contact and audio alongside millimeter-level positioning data.
The corporate says its testing signifies that 1,000 body-free knowledge samples mixed with 100 samples collected from actual robots can present coaching outcomes corresponding to roughly 1,000 real-robot samples.
X Sq. Robotic additionally claims its strategy could make knowledge assortment 2.33 instances extra environment friendly than standard remote-control strategies.
On the software program facet, QUANXTA Zero covers the workflow from assortment and cleansing by high quality management, annotation, mannequin coaching, simulation, analysis and mannequin iteration.
The intention is to show collected demonstrations into structured, reusable coaching property that may assist the persevering with growth of embodied AI fashions.
Taken collectively, the WRC demonstrations current the totally different layers of X Sq. Robotic’s technique: knowledge assortment and mannequin growth at one finish, and bodily robots performing industrial, dexterous and home duties on the different.
Logistics gives a real-world take a look at
The logistics demonstration carried out instantly earlier than WRC gives a extra production-oriented instance of how these applied sciences will be utilized.
Parcel induction is a very demanding automation downside as a result of packages arriving from unloading operations should not essentially introduced in predictable positions.
Packing containers overlap, comfortable packages deform, labels can face within the mistaken route and the obtainable greedy floor modifications each time a parcel is eliminated.
X Sq. Robotic’s strategy makes use of WALL-B to interpret these altering circumstances and decide how its six-axis arms ought to strategy and manipulate particular person parcels.
The arms can choose, flip and reposition packages earlier than feeding them right into a downstream conveyor. They’ll additionally flatten labels on versatile packaging and alter the orientation of bins to enhance subsequent barcode scanning.
In the course of the livestream, the system additionally demonstrated exception dealing with when an arm intervened to get better a parcel transferring towards the mistaken routing space.
Wang Qian, founder and CEO of X Sq. Robotic, says: “The query in logistics automation just isn’t whether or not a robotic could make one clear choose. It’s whether or not it retains making good choices because the pile modifications, recovers when one thing goes mistaken, and retains the remainder of the operation transferring. We designed the automation across the work itself, not round an idealized setting.”
That adaptability attracted consideration from automation specialists following the demonstration.
Katy Lin, an automation engineering supervisor within the automotive business, says: “Velocity will get the headline, however dealing with real-world variability at that velocity is what makes this spectacular.”
Function-built robots and the humanoid query
The livestream additionally attracted consideration as a result of X Sq. Robotic processed 1,816 parcels within the hour after setting itself a goal of 1,248 parcels per hour – a determine corresponding to the sustained hourly common achieved throughout Determine AI’s for much longer 200-hour humanoid logistics demonstration.
X Sq. Robotic exceeded that concentrate on by roughly 45 %.
The demonstrations should not instantly equal. Determine used a whole humanoid robotic and demonstrated endurance over a for much longer interval, whereas X Sq. Robotic’s system makes use of stationary robotic arms optimized for a parcel-induction workflow.
However the comparability raises a broader query that’s significantly related to X Sq. Robotic’s WRC portfolio: whether or not embodied intelligence must be related to one common robotic type.
Scarlett Lu, a China expertise and provide chain analyst, says: “That is the comparability I’d wish to see rather more usually. Not ‘robotic vs human’, however one robotic structure vs one other for a similar job.”
She provides: “Typically the successful robotic might merely be the one designed across the work, not across the human type.”
Throughput alone won’t decide which structure finally makes business sense.
Damijan Zorko, a specialist in gear transmissions and tribology, says: “1,816 vs. 1,248 parcels per hour is attention-grabbing, however what does every system price to buy, function and preserve?”
He provides: “Finally, the enterprise case will.”
Tim Schmiedl, co-founder and CIO of fruitcore robotics, summarizes the broader argument: “The shape issue is secondary – what issues is whether or not the system has the best bodily capabilities and the intelligence to orchestrate them reliably.”
That statement additionally brings the assorted applied sciences X Sq. Robotic is exhibiting at WRC collectively.
Six-axis logistics arms, dexterous fingers and robots working in properties symbolize very totally different bodily methods and purposes. X Sq. Robotic’s proposition is that basis fashions, scalable knowledge infrastructure and applicable robotic {hardware} will be mixed based on the setting and process.
Its WRC 2026 showcase gives a chance to display that proposition throughout the corporate’s portfolio – with the current logistics livestream providing one instance of what occurs when the expertise is utilized to a demanding real-world workflow.
