Robotics Executives Clash Over Mass Production Hurdles at WRC 2026

The executives split into two camps. According to Raynews, Zhao Tongyang, CEO of Zhongqing Robotics, said current embodied AI models achieve only about 96% success in operations, meaning four out of 100 tasks fail, which is unacceptable for industrial clients. He called for raising reliability from 99% to 99.9% or higher. Cheng Hao, CEO of Accelerated Evolution, agreed that the technology is not ready, describing the industry as being in the "dial-up internet era" of embodied intelligence, with no end-to-end large model having passed engineering tests.

However, Liu Yulong, co-founder of Jushen Technology, which delivered about 15,000 robots including around 10,000 quadruped robots by June 2026, argued that the real bottleneck is hardware consistency and manufacturing standards. He cited payload-to-weight ratio, mean time between failures (MTBF), and a lack of industry-wide standards. Jiao Jichao, vice president of UBTech, said customers care more about return on investment than extreme reliability, and raised the "cross-embodiment migration" problem, where models trained on one robot fail on another.

Xu Lei, head of JD.com's intelligent robotics business, said the biggest bottleneck is matching supply with demand, noting that the conversion rate for embodied intelligent robots is "dozens of times" worse than for traditional consumer electronics. The panelists also debated the industry's future structure, with Zhao and Jiao favoring vertically integrated giants, while Cheng argued that software and hardware will eventually decouple, leading to a super-large-model era.

In a separate WRC 2026 event, the "Physical AI Leaders Forum" took place on August 19, organized by Kuawei Intelligence and co-organized by Caijing magazine. As reported by QbitAI, Mei Tao, founder and CEO of HiDream.ai, said that while the IQ of top AI models has climbed from around 130 to close to 140 this year, high IQ does not guarantee reliable physical world performance. He announced the release of HiDream-O1-World, a native all-modal interactive world model based on the UiT architecture, which topped the Navi sub-ranking of the WBench benchmark. The model supports text, image, and interactive control inputs and maintains long-term spatiotemporal consistency, enabling AI interactive games, embodied simulation, and 3D scene generation.

Mei emphasized that world models are the cognitive core connecting data, models, agents, and robotic bodies. HiDream.ai has accumulated nearly 10 million hours of video data, and is partnering with Noitom to enhance human motion data generation for robot training. He described a closed loop where real-world feedback becomes new training data, which is essential for scaling embodied intelligence.

Separately, Ant Group announced in a release carried by QbitAI that the AFAC 2026 financial intelligent innovation competition will hold its final roadshow on August 25-26 at the Anadi Hotel in Shanghai. The event features roundtables, keynotes, and a comedy performance, with livestreaming. Since its first edition in 2023, the competition has attracted over 15,000 teams and nearly 50,000 participants, covering 600 universities and 400 enterprises, according to the organizers.