Two-Story Robot and Improved Demo Highlight 2026 World Robot Conference

According to Chinese tech media QbitAI, ZERO drew large crowds and became the most-watched exhibit on the opening day at the Yichuang International Convention Center. Designed by artist Zhu Gaowen, the robot adopts a minimalist style with Eastern aesthetic elements. Zhongjian said the giant scale is intended to prompt visitors to reconsider the relationship between humans and machines. The company deliberately left ZERO without a predetermined identity, describing it as "an ideological enlightenment under the AI cognitive paradigm."

Beneath the showpiece, Zhongjian displayed self-developed hub motors and planetary joint modules in transparent "disassembly compartments." ZERO is based on NVIDIA's Isaac open robot development platform for multi-scenario pre-training, while the Lingrui P1 quadruped robot runs on the NVIDIA Jetson Orin edge computing platform. The company said its in-house factory can produce more than 1,000 such units per month. It also presented the commercialized industrial-grade Lingrui P1 for grid inspection and park patrol, and a firefighting robot dog based on the P1 platform. The firefighting version carries a water cannon with a maximum range of 60 meters and can switch between a straight water column and a 120-degree wide-angle spray. It also has an automatic water hose release system for emergency evacuation. Zhongjian said these products rely on a full-stack self-developed chain of perception hardware, visual algorithms, and edge computing.

Zhongjian has also invested in the humanoid robot company 1X Holding AS. The company, founded in 1997 and listed in Shenzhen since 2015, officially started its embodied intelligence strategy in 2023. Its R&D expenses for the first quarter of 2026 jumped 291.23% year-on-year. The company is also exploring consumer robots, hoping to bring industrial-grade motion control and safety standards into homes.

Another exhibitor, Qianxun Intelligent, brought the same "tidy the living room" demo it showed at WAIC a month earlier, but with quantified improvements. According to QbitAI, the success rate for placing a can of cola into a refrigerator rose from about 80% to over 99%, and for placing dishes into a dishwasher from about 90% to over 99%. The time to confirm the navigation target dropped by nearly 50%. Under randomized conditions—varying colors, types, and positions—the trash-picking task achieved an 85% success rate.

Qianxun attributes these gains to what it calls "time density," a measure of how quickly a system can absorb failures, retrain, and re-deploy. Its data platform collects web videos, first-person views, teleoperation data, and real-world task data. The company said its self-developed data acquisition equipment has been upgraded to the seventh generation, cutting collection costs to one-tenth of traditional teleoperation and raising data availability from 30% to 95%, with more than 300,000 collection points nationwide. The company promotes "dirty data" training to make models robust to real-world variations such as changing light and moving people.

Its foundation model, Spirit v1.6, integrates vision, language, action, and world-state prediction into a unified architecture. According to QbitAI, the model ranked first in the RoboArena international benchmark launched by UC Berkeley, Stanford, and NVIDIA. An agent layer maintains task state and handles failures by deciding whether to continue, retry, skip, or replan. Navigation now defines arrival as reaching an "operable pose" rather than a mere coordinate. Optimizations reduced the median time to generate the first navigation target by nearly 50%. Hardware engineers also shifted to front-end design reviews to prevent issues like servo failures and loose connectors.

In brand-new settings at WRC, Qianxun achieved 100% success on several core tasks through zero-shot transfer. The team continues to stress-test by moving starting positions, altering object poses, and changing backgrounds. The company's Moz1 robot has entered CATL's power battery PACK production line, where it handles non-standard processes such as plugging high-voltage test connectors, with a success rate above 99% and daily workload equivalent to three times that of a skilled worker. Qianxun is also cooperating with JD.com, Bosch, and Schaeffler.