Mifeng Technology Launches Crowdsourced Data Platform for Robot Training
On the afternoon of Sept. 23, Mifeng Technology officially launched “Mifeng Pai”, an embodied data crowdsourcing collection platform, at an event in Shanghai. The platform consists of the company’s self-developed MEgo collection devices, the Mifeng Pai app and the MEgo Engine data governance engine, and organizes data production for model makers and robot companies. The company also introduced the concept of “robot trainer” as a new occupation.
According to Mifeng, the platform covers 22 scenario categories, more than 5,000 tasks and more than 50,000 real-world environments, spanning logistics, warehousing, accommodation, catering, retail, elderly care, medical care, agriculture, intangible cultural heritage and other fields. During a month-long beta test before the official launch, the platform registered 20,000 users and received 13,000 task submissions.
Yao Maoqing, chairman and CEO of Mifeng Technology and a co-founder of embodied intelligence company Zhiyuan Robot, said the team aims to capture the manipulation actions of ordinary people in their daily work and turn them into training data for robots. Unlike teleoperation with physical robots, which depends on robot availability, the platform uses “body-free” collection: it first records human operations, then reconstructs trajectories and maps actions to fit different models and robots. Users can accept tasks in the app, wear MEgo devices to perform specified operations in scenarios such as retail, warehousing, manufacturing and homes, and get paid based on approved effective duration. MEgo Engine then handles data processing, annotation, quality assessment and delivery, decomposing operations into more than 120 standard atomic actions such as grasping, placing, releasing, twisting and pressing.
Yao said current data demand has reached millions of hours, and crowdsourcing is more flexible for expanding into professional skills such as pipe repair, car repair, hairdressing and nail art. The company said data that passes its settlement process has an availability rate of more than 95 percent, and its million-hour body-free dataset has begun serving clients including Tencent Robotics X Lab and Ant Lingbo.
