On March 30, 2026, in Shenzhen, a quiet however telling milestone passed off within the embodied AI world. X Sq. Robotic (自变量机器人) hosted the inaugural Embodied AI Builders Convention — EAIDC 2026 — billed because the world's first large-scale gathering devoted particularly to builders constructing embodied AI methods. The occasion introduced collectively researchers, engineers, and expertise corporations from throughout the trade for stay robotic demonstrations, a national-level hackathon, and centered discussions on a subject that the majority robotics conferences nonetheless deal with as futuristic: not whether or not robots might be clever, however find out how to deploy them at scale in the true world.
That framing — from can it work to how can we ship it — is exactly what distinguishes X Sq. from most corporations on this house. Whereas rivals have spent the previous two years perfecting demo movies of robots doing backflips, X Sq. has been obsessively centered on a more durable, much less photogenic downside: constructing a basis mannequin basic sufficient to really run in an actual residence atmosphere, a messy and variable house with out failing the second one thing surprising occurs.
EAIDC was, in some methods, a coming-out get together. X Sq. is now not a well-funded startup working within the background. It’s more and more the corporate setting the phrases of the dialog. Following the conclusion of the convention, Wang Qian(王潜), X Sq.'s founder and CEO, sat down with Pandaily for an unique interview to debate the street forward.
What X Sq. Truly Builds
To know X Sq., it helps to know what downside it’s making an attempt to unravel.
Conventional industrial robots are basically very exact, very quick machines that observe pre-programmed directions. They excel in structured environments — the identical process, the identical object, the identical place, repeated 1000’s of instances. Change any variable they usually fail. They don’t have any skill to look at the world, purpose about it, and adapt.
The corporate's reply is what it calls an Embodied Intelligence Basis Mannequin — particularly, a Imaginative and prescient-Language-Motion (VLA) mannequin that processes sensory enter (video, language, haptic alerts) and outputs bodily motion (joint torques, velocities, poses) in a single end-to-end structure. The mannequin, branded the Nice Wall sequence with its flagship referred to as WALL-A, is designed to allow robots to understand, resolve, and act autonomously throughout numerous environments and duties with out being explicitly reprogrammed for every one.
X Sq. claims WALL-A is at present one of many world's largest-scale unified embodied basis fashions — achieved through a novel three-stage coaching paradigm, a proprietary structure combining shared consideration with expert-routed feed-forward modules, and coaching information drawn from a mixture of self-collected real-world demonstrations, open-source robotic datasets, and routinely generated multimodal information.
In September 2025, X Sq. open-sourced a developer-facing model of its mannequin referred to as WALL-OSS, which has since been built-in into Hugging Face's LeRobot framework — a transfer that concurrently builds neighborhood goodwill and feeds the information flywheel. EAIDC itself might be learn as an extension of this technique: by convening the ecosystem, X Sq. is positioning its open-source mannequin because the platform builders construct on prime of.
On the {hardware} facet, X Sq. produces two robotic platforms: Quanta X1, a wheeled dual-arm robotic designed for precision manipulation in service environments, and Quanta X2, a wheeled humanoid launched in August 2025. It additionally makes the Artixon Hand, a 20 DoF dexterous robotic hand — which the corporate claims is the world's first such hand to be managed straight by a basis mannequin somewhat than conventional motion-planning algorithms.
The corporate is headquartered in Shenzhen with places of work in Beijing and Shanghai, and has deployed its robots in lodge teams, logistics corporations, aged care services, supermarkets, and analysis establishments. Early industrial income is coming in from schooling, hospitality, and elder care, with family companies — together with a collaboration with classifieds platform 58.com — at present energetic.
The Founder: A Transformer Pioneer Who Stop Finance to Construct Robots
Wang Qian is just not a typical robotics entrepreneur. His educational background is in deep studying — particularly, he was among the many earliest researchers on this planet to introduce the eye mechanism into neural networks, publishing work on the identical convention as Google's early consideration paper in 2014, three years earlier than the Transformer structure would reshape all the subject of AI.
After finishing his bachelor's and grasp's levels at Tsinghua College, he pursued a PhD on the College of Southern California, conducting robotics studying and human-robot interplay analysis at prime American robotics laboratories. However he pivoted mid-career — to quantitative finance. He based a quant fund in the US, which by his personal account was financially profitable. After which he spent many nights unable to sleep.
"I stored pondering: I ought to have stayed in robotics," Wang has mentioned in interviews. In 2023, he dissolved the fund and returned to China.
His co-founder and CTO, Wang Hao (王昊), brings a complementary profile: a PhD in computational physics from Peking College, he beforehand led the algorithm group for the Fengshenbang massive language mannequin on the Institute for Clever Computing (IDEA Analysis), overseeing improvement of China's first hundred-billion-parameter basis mannequin and one of many earliest trillion-parameter fashions, Ziya.
Collectively, the 2 founders embody X Sq.'s core thesis: that constructing a general-purpose robotic mind requires experience in each massive basis fashions and robotics studying — and that the 2 can’t be solved in isolation.
"We Had been the Solely Variable"
The title "X Sq." is just not arbitrary. In arithmetic, x represents the unbiased variable — the factor that modifications and drives outcomes. The Chinese language title, 自变量 (zì biànliàng), carries the identical which means with an extra nuance: zì means "self" or "autonomous." The corporate needs to be the variable that modifications the world — and it needs that change to be self-generated, not depending on others.
This framing seems to be an correct description of how X Sq. navigated its early years.
When the corporate was based in late 2023, the embodied intelligence panorama in China was already aggressive. Galbot (银河通用) and Agibot (智元机器人) had each launched that very same yr with appreciable fanfare, bigger groups, and extra preliminary funding. X Sq., against this, operated with little public visibility and struggled to boost early rounds.
"The largest issue was that no one trusted us," Wang Qian has recalled. "Early buyers didn't imagine we might really pull this off. They guess on the group being respectable sufficient that even when we failed at this, we'd discover one thing else to do."
What modified was output. X Sq. launched its first embodied intelligence mannequin simply two months after founding. By October 2024, it had skilled WALL-A — on the time, the largest-parameter general-purpose embodied manipulation mannequin on this planet. The demos confirmed one thing qualitatively completely different: robots that would hold laundry, put together shaved ice, wind cables round pegs, and kind parcels of arbitrary form, all managed by a single mannequin with out task-specific reprogramming. By the point Bodily Intelligence (PI), the American benchmark firm on this house, launched its π0 mannequin in late 2024 — validating the end-to-end VLA strategy — X Sq. had already been on that path for over a yr.
"We didn't want to repeat anybody's homework," Wang has mentioned. "We had been already there."
By early 2025, the funding image had remodeled completely. Meituan led a Sequence A spherical; Alibaba Cloud and Hongshan (previously Sequoia China) co-led an A+ spherical of almost RMB 1 billion; and by January 2026, a 1 billion yuan A++ spherical closed with ByteDance, Hongshan. The corporate has now raised roughly $280 million in whole — and is the one Chinese language embodied intelligence startup to have concurrently attracted backing from Meituan, Alibaba, and ByteDance.
The Technical Guess: Why Finish-to-Finish Issues
The central debate in embodied intelligence — and the one X Sq. has wagered its existence on — is architectural. Ought to a robotic's "mind" be constructed as a modular stack (notion module → planning module → management module), or as a single end-to-end mannequin that processes uncooked sensor enter and outputs uncooked motor instructions with out express intermediate representations?
X Sq.'s reply has been unambiguous from day one: end-to-end, all the time.
Wang Qian's reasoning is each technical and philosophical. Modular methods, he argues, impose human-designed abstractions onto a steady bodily world. Every handoff between modules introduces error, latency, and brittleness. The bodily world doesn't phase neatly into "notion" and "planning" — it’s a steady stream of drive, contact, spatial relationships, and temporal dependencies. A mannequin that learns to navigate that stream as a complete, somewhat than as a sequence of sub-problems, will generalize higher and fail extra gracefully.
"Embodied intelligence is a basis mannequin for the bodily world — unbiased from, and parallel to, language fashions and multimodal fashions for the digital world," Wang mentioned on the 2025 MEET convention. "The long-term development of robotics intelligence will depend on general-purpose AI functionality."
The sensible implication is that WALL-A takes language, video, and haptic sensor information as enter and outputs velocity, pose, and torque straight — with no intermediate planning layer. The identical mannequin controls the corporate's wheeled robots, its dexterous hand, and in precept any robotic platform it’s skilled on. In July 2025, X Sq. demonstrated what it claims was an trade first: a basis mannequin controlling a high-degree-of-freedom dexterous hand to carry out advanced manipulation duties, together with card dealing, with out task-specific engineering. The corporate has additionally built-in chain-of-thought reasoning natively into the mannequin — enabling robots to interrupt down advanced directions into subtask sequences and purpose by them earlier than performing.
The place the Robots Are Truly Working
For all of the mannequin functionality claims, X Sq. has been notably centered on real-world deployment — and particularly on environments which might be messy, variable, and require real generalization.
The corporate's most commercially developed vertical is hospitality. It has deployed robots at lodge teams throughout China, the place the Quanta X1 and X2 platforms deal with visitor reception, baggage help, laundry — together with the complete cycle from assortment to washing to folding and return — room supply, and cleansing. The corporate has additionally deployed at aged care services, the place robots carry out transport, monitoring, and companion duties across the clock.
In logistics, X Sq. has deployed at main home parcel sorting corporations, the place its reinforcement studying integration permits robots to type parcels of numerous shapes at speeds the corporate says exceed human efficiency. In retail, it has partnered with grocery store chains the place Quanta X1 performs restocking duties autonomously.
This yr, X Sq. is doubling down on deploying robots in residence environments—a much more advanced house outlined by higher process variability. In contrast to managed settings, houses current actual kitchens, actual residing rooms, and actual bedrooms, every with distinctive layouts, objects, and consumer habits. A single process like "fetch a cup" performs out in another way throughout 100 households. True generalization, the corporate argues, can solely be achieved by coaching on real-world information. In spite of everything, "Regardless of how lengthy you prepare in a pool, you gained't study to swim within the ocean."
These deployments matter past income. They’re X Sq.'s information engine. The corporate has invested closely in proprietary teleoperation and information assortment {hardware} — together with exoskeleton-based information seize methods — to generate the high-quality real-world demonstration information that trains WALL-A. Wang Qian has described this closed loop of {hardware}, information, and mannequin as X Sq.'s major aggressive moat.
"The following section of competitors in embodied intelligence," he has mentioned, "is actually a battle of basis fashions constructed on information closed-loops and their capability for mannequin evolution."
EAIDC, considered by this lens, can be a knowledge technique. By constructing a developer ecosystem and internet hosting a hackathon, X Sq. is making a funnel of contributors who lengthen the mannequin's protection into new duties and environments — with out X Sq. having to gather each information level itself.
Wang Qian's Worldview: Lengthy-Time period Believer, Quick-Time period Skeptic
For all his confidence in X Sq.'s personal trajectory, Wang Qian is measured — typically contrarian — when he talks in regards to the broader robotics trade.
On the lengthy arc, he’s a real believer. He thinks robots will likely be ubiquitous inside a decade, dealing with each class of bodily labor that people at present carry out. He has spoken about robots finally engaged on Mars. The query for him is never_ whether_ — it’s when, and who will get the structure proper.
On the close to time period, he’s extra cautious. He thinks numerous high-profile robotics corporations are over-promising and over-valued, and that the trade would profit from much less hype and extra sensible expectations about what the present technology of fashions can really do reliably at scale. His view is that the sector is roughly on the equal of GPT-2 within the massive language mannequin improvement curve — genuinely spectacular, unmistakably the suitable course, however nonetheless nicely wanting the inflection level that unlocks mass-market utility.
On the place robots ought to be deployed first, Wang Qian has a transparent hierarchy. Service industries — resorts, aged care, property administration — come earlier than factories. The reason being partly industrial (service environments reward generalization; factories reward repetition) and partly technical. Manufacturing unit work, he argues, calls for extraordinarily excessive precision and throughput speeds that present embodied fashions can’t constantly ship, whereas providing restricted information variety in return. Service environments, against this, are wealthy in variation, which is precisely what trains a greater basic mannequin. "Working in a manufacturing unit doing easy, repetitive duties — that's principally a PR transfer," he mentioned.
On the query of aggressive moats, Wang Qian believes {hardware} will matter extra over time than most software-first buyers at present assume. {Hardware} is sticky: switching prices are excessive, customers develop preferences, and bodily type elements create model loyalty in ways in which pure software program doesn’t. However he additionally thinks the post-ChatGPT period has basically elevated the significance of mannequin functionality — and that corporations which under-invest in basis mannequin analysis will discover their {hardware} benefits eroded shortly. The successful system, in his view, is the one X Sq. is pursuing: {hardware} and mannequin co-developed from the bottom up, with every informing the opposite.
From Product Definition to Customary Setting: X Sq.’s World Management
Wang Qian’s confidence stems not solely from his confirmed enterprise logic and a high-execution group but additionally from the uncommon distinction of securing backing from Meituan, Alibaba, and ByteDance—three giants that seldom align on investments. Now, because the host of EAIDC, Wang is transitioning X Sq. from a product firm right into a convener for the subsequent technology of embodied AI. At CVPR 2026, X Sq. Robotic partnered with Solar Yat-sen College and MBZUAI to launch ManipArena, a platform establishing a unified world benchmark with 20 real-world duties and 188 hours of high-quality information. By providing cloud-based distant testing, X Sq. is decreasing {hardware} obstacles and defining the foundations for the trade's subsequent section: constructing a world, open, and extremely collaborative developer ecosystem. As Wang famous, "I don't must persuade anybody; the outcomes converse."
Elevate your perspective with NextTech Information, the place innovation meets perception.
Uncover the most recent breakthroughs, get unique updates, and join with a world community of future-focused thinkers.
Unlock tomorrow’s traits right this moment: learn extra, subscribe to our e-newsletter, and change into a part of the NextTech neighborhood at NextTech-news.com
