MANUS™(@ManusMeta):New study from @Stanford , @MIT , and @scale_AI , "Pre-training Visual Dexterity in Simulation," pre-trained a bimanual robot policy on 75 hours of simulated demonstrations, then fine-tuned it with just one to two hours of real data per task. For the fine-tuning stage, operators wore MANUS gloves for finger tracking paired with a Quest controller for wrist tracking, producing the smoother, more precise control needed to guide the robot through occlusions and at a distance. Read more:

MANUS™

@ManusMeta

Building scalable solutions for human interaction data. Join our team:

加入 June 2014

352 正在關注 4.5K 粉絲

MANUS™@ManusMeta

2026.08.19 10:49

New study from @Stanford , @MIT , and @scale_AI , "Pre-training Visual Dexterity in Simulation," pre-trained a bimanual robot policy on 75 hours of simulated demonstrations, then fine-tuned it with just one to two hours of real data per task. For the fine-tuning stage, operators wore MANUS gloves for finger tracking paired with a Quest controller for wrist tracking, producing the smoother, more precise control needed to guide the robot through occlusions and at a distance. Read more: