Fine-grained Manipulation

Manipulation with tight tolerances for position and force, such as inserting a battery, threading a wire, or turning a tiny screw.

Fine-grained manipulation refers to tasks with very little room for error, requiring millimeter-level positioning or careful force control — inserting a battery into a slot, peeling the lid off a small translucent cup, threading a wire through a small hole. There are two main difficulties: visually, the target is small and often hidden by the hand or gripper, so even a slight misalignment causes failure; on contact, applying even slightly too much force crushes the object or jams the mechanism. Early approaches relied on precise modeling plus force control; end-to-end imitation learning is now the dominant approach. Tony Zhao and colleagues' 2023 ALOHA paper used a low-cost bimanual setup with the ACT algorithm to reach 80–90% success on six fine-grained tasks with only about 10 minutes of demonstrations each. Other work brings in touch, such as 3D-ViTac, which fuses vision and touch to handle fragile objects. This term often appears alongside dexterous manipulation, but dexterous manipulation puts more emphasis on multi-fingered hands specifically.

ExampleALOHA's two arms peel the lid off a small translucent sauce cup, or insert a battery precisely into its slot.

Also called

Precise Manipulation, Fine Manipulation

Related

Dexterous Manipulation · Contact-rich Manipulation · Action Chunking with Transformers · 3D-ViTac · Visuo-Tactile Fusion · Bimanual Manipulation

Sources

Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware (ALOHA / ACT, arXiv 2304.13705)
3D-ViTac: Learning Fine-Grained Manipulation with Visuo-Tactile Sensing (arXiv 2410.24091)