X Square Robot自变量机器人
LatestSeries C (B/B+/B++/C rounds all closed by Jun 2026)
Above RMB 20B (the first and only embodied AI company in the Greater Bay Area to break RMB 20B)
X Square Robot Technology (Shenzhen) Co., Ltd.
The only embodied AI company to have gathered investments from ByteDance, Alibaba Cloud, Meituan, Xiaomi, China Mobile and other giants; WALL-B world unified model; the 1,816 items/hour sorting world record surpassing Figure AI at only ~30% of the cost
数据截至2026-08-19
About
X Square Robot was founded in Shenzhen in December 2023; founder Wang Qian holds BS/MS degrees from Tsinghua and a PhD in robotics from USC, and was among the earliest researchers to introduce attention mechanisms into neural networks. The company is a representative Chinese general embodied model company betting on the 'fully end-to-end' route; in June 2026 it closed its B/B+/B++/C rounds in full, with cumulative funding over RMB 4B and a valuation breaking RMB 20B — the first embodied AI company in the Greater Bay Area above RMB 20B, and the only one in China simultaneously invested in by ByteDance, Meituan, Alibaba Cloud and Xiaomi (with China Mobile added in 2026). Its models evolved from WALL-A and the open-source WALL-OSS (No. 2 globally) to WALL-B, released in April 2026 as the world's first World Unified Model (WUM) architecture, jointly training vision, language, touch, action and physical prediction in a single network. Its robot product is the Liangzi 2 general wheeled humanlike robot (62 DoF). Commercially, in August 2026 a dual-arm + standard-gripper logistics sorting solution demonstrated 1,816 items/hour at 98%+ accuracy in a live-streamed real test, surpassing Figure AI by about 45% at only ~30% of the cost — called embodied AI's 'DeepSeek moment'; with 58 Daojia it launched a smart housekeeping service, becoming the world's first company to bring embodied robots into ordinary households at scale.
Technical Approach
'Unified brain-cerebellum': instead of a layered big-brain/cerebellum design, a single unified end-to-end VLA / world unified model handles perception, decision-making and execution; insists that 'embodied foundation models cannot copy foreign homework' and pre-trains original models from scratch; model-driven hardware reduction — push complexity into the brain and simplify the body (in logistics, dual arms + standard grippers beat humanoid solutions on efficiency); the same underlying model is reused across scenarios, flowing from homes to warehouses and later to manufacturing and retail.
Commercialization（6）
- On August 12, 2026, completed an about-1-hour fully autonomous live-streamed logistics sorting test: a dual-arm + standard-gripper setup faced real parcels of random size/shape/material/weight with no human intervention, achieving 1,816 valid items/hour at 98%+ accuracy
- Efficiency about 45% above Figure AI's published 1,248 items/hour ('145% efficiency at 30% cost'), called embodied AI's 'DeepSeek moment' by media; hardware cost down about 70%, with the whole system costing about 30% of comparable U.S. solutions
- Signed cooperation with a leading domestic logistics company; the equipment has run in real scenarios for months and is expanding; less than three months from R&D to a working solution
- With 58 Daojia, launched a directly bookable smart housekeeping service: a robot + cleaner visit homes together, completing living-room tidying and other chores in about 3 hours; WALL-B is being validated and iterated in real home scenarios — the world's first company to bring embodied robots into ordinary households at scale
- Formed joint ventures with Jinbei Auto (April 2026) and Jack Technology (apparel industry), and cooperates with 沈汽集团 (auto production)
- Liangzi 2 has opened recruitment for entry into real homes
Position & Strengths（5）
- Top-tier capital lineup: ByteDance + Meituan + Alibaba Cloud + Xiaomi + China Mobile + HongShan + SAIC; cumulative funding over RMB 4B and valuation above RMB 20B
- Forward-looking technical route: WALL-B, the world's first World Unified Model (WUM) architecture, breaks through the VLA module-stitching ceiling
- Open-source ecosystem influence: WALL-OSS ranked No. 2 globally; WALL-OSS-0.5 goes straight onto real robots after pre-training; the 'Juliang Plan 2026' hackathon
- Dual benchmark in efficiency and cost: 1,816 items/hour surpassing Figure AI by 45% at 30% cost, proving 'if the model is smart enough, the body can be simplified'
- Generality validated: the same WALL-B reused across home → logistics → manufacturing scenarios