Naive AI Raises $400 Million in Seven Months as Chinese Vision Talent Moves Into Robotics

The report said Naive AI's website carries only an English slogan, with no team photo or product introduction, and the team has not given interviews. Dai's founding group is described as eight people: Dai, Zhu Xizhou, Sun Peng, Chen Yuntao, Wang Wenhai, Chen Zhe, Luo Gen and Wang Weiyun. The report traces their links through Microsoft Research Asia, SenseTime, Shanghai AI Laboratory and Tsinghua.

Zhu Xizhou, a joint Ph.D. student between the University of Science and Technology of China and Microsoft Research Asia, worked with Dai on Deep Feature Flow, which combined optical flow and video recognition end to end and sped up video recognition by five times, or up to ten times with a lighter optical-flow network, according to the report. Zhu was first author of Deformable DETR, a paper cited 12,000 times on Google Scholar. Sun Peng, a Nanyang Technological University Ph.D. and AI infrastructure specialist at SenseTime, led work that used 512 Nvidia Tesla V100 GPUs to train AlexNet on ImageNet in 90 seconds, a 410-fold speedup, the report said.

Chen Yuntao, a Ph.D. from the Institute of Automation of the Chinese Academy of Sciences, worked on autonomous-driving perception at TuSimple and Huixi Intelligent. He was corresponding author of DrivingDojo, described by Leiphone.com as the first dataset designed for interactive world models, with 18,000 videos, accepted by NeurIPS 2024. Wang Wenhai was first author of InternImage, a CVPR 2023 Highlight paper that reached 65.4 on COCO. Chen Zhe was first author of InternVL, a CVPR 2024 Oral paper. Luo Gen, a postdoctoral researcher from Xiamen University, worked on Mono-InternVL. Wang Weiyun, born in 1999, is a joint Ph.D. student at Fudan University and Shanghai AI Laboratory and first author of InternVL 3.5, whose InternVL-Chat passed 10 million downloads on HuggingFace and scored 77.7 on MMMU, according to the report.

In a separate Leiphone.com report, Benewake founder and CEO Li Yuan said the company delivered 250,000 robot lidar units in the first half of this year. Robot business accounted for more than half of revenue, and revenue from that business doubled in 2025 from 2024, the report said. Li, a student of Xu Xurong, a founder of Chinese luminescence studies, founded Benewake in 2015 after returning to China. The company describes itself as a Robot Native lidar maker; Li said robotics is now its main battlefield and automotive lidar was a capability migration.

Benewake's products include the UW500 for underwater use, the Horn series for rail transit and the NOVA20 for embodied intelligence, which uses a 'slow-thinking brain and fast-perceiving cerebellum' approach, according to the report. Li said defective-product rates are now at a few parts per million and annual deliveries are near one million units. He said the company's early customers included ZeroTech, then a drone maker, and that it first worked on drone, industrial and rail applications before entering automotive lidar around 2021.

Leiphone.com also recounted a 2002 horse race inside Microsoft Research Asia. Zhang Lei, then a young Ph.D. who had joined MSRA in 2001, led a small team including Xiao Rong, Zhu Long and Chen Longbin against Li Ziqing, whose Eye-CU face-recognition system had been demonstrated by Bill Gates and covered by CNN. Zhang redesigned the evaluation metric around false positives rather than detection rate, rewrote the inference code in C++ and added visualization. After three months, the false-positive rate fell from what the report described as 'buy one, get one free' to 1%, and the system was integrated into Windows.

The report said Zhang's interns and team members later included Huang Zheng, who founded Pinduoduo; Zhu Long, who founded Yitu Technology; Yan Shuicheng, later elected to the Singapore Academy of Engineering; Xiao Rong, now a technology VP at Intellifusion; Xu Pengfei, now CTO of Xin Jianzhi Robotics; and Wang Changhu, founder of Aishi Technology. Shen Xiangyang, then at MSRA and later Microsoft's global executive vice president, remembered Zhang after the race. Zhang's doctoral advisor, Zhang Bo, told him to 'understand the problem thoroughly,' a principle the report said shaped his career.