The realm of embodied AI has witnessed significant advancements in data acquisition, particularly with surface electromyography (sEMG) technology, which focuses on the output aspect of human movement and has become a key industry focus. OriginFlow, a startup founded by Tsinghua PhD student Qin Shentao, has captured significant investment interest with its NeuroScale technology system. Despite being established for less than a year, the company has successfully completed three rounds of financing, amassing over 500 million yuan.

The NeuroScale technology system seamlessly integrates neuromuscular electrophysiology, first-person visual data, and IMU sensor information to construct a comprehensive three-dimensional physical model. This innovative approach transforms human actions into machine-learnable representations known as ‘Human Tokens.’ Its hardware offering, the OriginKitGen 1.0 wristband, is capable of monitoring forearm muscle activity in real-time. At the algorithmic level, the PULSE foundation model efficiently converts raw waveforms into structured data. Furthermore, the team has developed a comprehensive Data Infra platform that covers the entire data processing pipeline, providing datasets that are directly applicable for robot training.

Currently, the technology is evolving from experimental wristbands to consumer-grade products, aiming for ‘unobtrusive’ data acquisition to achieve data accumulation on a trillion-hour scale. One of the standout features of this technology is its cross-agent transfer capability, which allows the adaptation of human operational experience to robots with varying configurations. This significantly enhances cross-agent task adaptation efficiency by over 40%.

Domestic companies such as BrainCo and Octopus Robotics are also exploring the neuromuscular electrophysiology field, contributing to the formation of a complete industrial chain and propelling embodied AI towards breakthroughs in precise physical interaction. Looking ahead, as acquisition devices transition to consumer-grade products, data derived from human daily behavior will emerge as a new driving force for the evolution of physical AI.