WRC 2026 Showcases Shift in Embodied AI: From Demos to Dependability and Dollars
At WRC 2026 in Beijing, embodied AI companies shifted focus from flashy demos to stable long-hour operation and viable business models, with logistics sorting and service scenarios leading the charge.
The shift was evident in the logistics sorting arena, where multiple companies showcased their solutions. Yuanli Lingji, using its newly released general-purpose foundation model DM0.5, sorted yellow packages from a mixed pile at a rate of about two seconds per item, roughly 1,200 pieces per hour. The same system also handled cutting cucumbers and stacking blocks, highlighting its generalization capability without task-specific training. In contrast, Xingdong Jiyuan impressed with its full-stack approach, using its dexterous hands to sort items on a live conveyor belt. A company representative told Leiphone that they welcome more players but noted that sorting is a scenario demanding deep operational experience. Zibian Robot emerged as a dark horse, prioritizing speed and efficiency. Its proprietary WALL-B model processes visual, language, and action inputs in a single network, enabling the robot to predict situations such as flipping a package based on subtle side indentation or clearing a jam only after managing an influx of parcels.
Beyond sorting, companies are expanding into office, retail, and household scenarios. UBTech showcased a fleet of a dozen Walker S2 humanoids working collaboratively under its "Group Brain Network 2.0," handling material handling and sorting autonomously, with 3-minute automatic battery swap supporting continuous operation. Galbot (Yinhe General) highlighted its brain-cerebellum architecture, with its AstraBrain-WBC 0.5 model achieving 0.39-millisecond inference latency for dynamic movements, while the system demonstrated long-horizon household tasks like folding shirts and making breakfast in a cluttered environment. Qianxun Intelligence showed its Mozi robot performing a "living room tidying" task, claiming a 15% improvement in efficiency over the past month, now autonomously adapting to changes like a moved laundry basket without needing new commands. Kuwei Intelligence displayed a single brain handling three diverse tasks—phone assembly inspection, supermarket scanning and sorting, and consecutive ice cream and popcorn serving—alongside a real industrial case with lens maker Lens Technology. Wudong Power brought a café to the expo, where its K15 robots served customers in a live environment, backed by a cumulative order volume of 700 million yuan and initial global deliveries.
The focus on generalization and operational stability reflects a broader industry consensus. Han Lei, head of world models at Tsinghua-affiliated home-robot firm Qiuzhi Technology, told Leiphone, "Debates over world-model definitions are just rhetoric; we care only about generalization." Fan Haoqiang of Yuanli Lingji echoed similar thoughts, saying the firm strives to achieve better generalization than π0.5. As the WRC demonstrations made clear, the embodied AI sector is moving beyond showcasing potential to delivering dependable, cost-effective solutions in real-world settings.