I\u2019m trying to understand what robotics and embodied AI teams actually need when collecting real-world training data.<\/p>\n
The use cases I keep hearing about are:<\/p>\n
-robotic hand manipulation<\/p>\n
-grasping and pick-and-place<\/p>\n
-soft and fragile object handling<\/p>\n
-tabletop tasks<\/p>\n
-warehouse tasks <\/p>\n
For teams working on imitation learning, VLA models, or robot manipulation, what is usually the biggest bottleneck?<\/p>\n
-not enough real-world data<\/p>\n
-task diversity<\/p>\n
-camera and sensor consistency<\/p>\n
-annotation quality<\/p>\n
-hardware-specific data<\/p>\n
I work with a small team connected to robotic visual data collection, but I\u2019m mainly trying to understand what teams actually need before going too deep in the wrong direction.<\/p>\n<\/div>\n
submitted by \/u\/WideAmbition1964 <\/a>
[link]<\/a><\/span> [comments]<\/a><\/span><\/p>
