给机器人当老师，还能赚外快？“中国版Index”觅蜂派来了
Ranking
Overall
61
Content
65
Popularity
N/A
No observed public metrics; popularity remains neutral/archived.
Merged summary
TL;DR - Mifeng Technology launched Mifeng Pai, a crowdsourced platform that pays people to collect real-world human activity data for training robots. Its integrated capture hardware, data-processing engine, and contributor network aim to address embodied AI’s growing need for diverse, large-scale training data.
- Mifeng says 20,000 MEgo wearable or gripper-based devices have collected over one million hours of data across 22 scenario categories, including images, depth, motion, touch, and audio.
- Its MEgo Engine reconstructs hand, body, device, and environment motion; the company claims sub-centimeter trajectory errors and reliable hand reconstruction even under severe occlusion.
- The ManiEval system grades rather than indiscriminately discards compliant data, matching precise demonstrations, failures, rare cases, and semantically rich recordings to different training and evaluation uses.
- Contributors receive equipment and payment for approved collection tasks; a pre-launch trial reportedly attracted 20,000 users and 13,000 submissions, while the resulting data is already serving customers including Tencent Robotics X and Ant Lingbo.
Sources (1)
给机器人当老师，还能赚外快？“中国版Index”觅蜂派来了
Public signals
N/A
TL;DR - Mifeng Technology launched Mifeng Pai, a crowdsourced platform that pays people to collect real-world human activity data for training robots. Its integrated capture hardware, data-processing engine, and contributor network aim to address embodied AI’s growing need for diverse, large-scale training data.
- Mifeng says 20,000 MEgo wearable or gripper-based devices have collected over one million hours of data across 22 scenario categories, including images, depth, motion, touch, and audio.
- Its MEgo Engine reconstructs hand, body, device, and environment motion; the company claims sub-centimeter trajectory errors and reliable hand reconstruction even under severe occlusion.
- The ManiEval system grades rather than indiscriminately discards compliant data, matching precise demonstrations, failures, rare cases, and semantically rich recordings to different training and evaluation uses.
- Contributors receive equipment and payment for approved collection tasks; a pre-launch trial reportedly attracted 20,000 users and 13,000 submissions, while the resulting data is already serving customers including Tencent Robotics X and Ant Lingbo.