Bimanual Manipulation

A robot uses two arms together, coordinating them to complete a single task.

Bimanual manipulation means a robot controls two arms — or, on a humanoid, two hands — at once to jointly carry out a task: holding a bottle steady with one hand while unscrewing the cap with the other, say, or folding clothes with both hands together. Many everyday tasks are impossible or clumsy to do with a single arm, which makes this a core capability for housework, assembly, and similar settings. The challenges are that the action dimensions double, the two arms must be coordinated precisely in both time and space, and the task often involves soft, deformable objects and dense contact. A landmark project is ALOHA (2023), from Tony Zhao, Chelsea Finn, and colleagues: a low-cost bimanual teleoperation platform paired with the ACT algorithm reached an 80–90% success rate on fine tasks like opening a translucent condiment jar or inserting a battery into a remote control, using only about 10 minutes (50 demonstrations) of data per task — though the hardest task, threading a zip tie, managed only 20%. Mobile ALOHA (2024) later mounted the same two arms on a mobile base.

ExampleWith just 20–50 human demonstrations per task, Mobile ALOHA learned to autonomously stir-fry shrimp and plate it, open a two-door cabinet and put away a pot, and even call and ride an elevator — tasks that need both bimanual coordination and mobility.

Also called

Dual-arm Manipulation

Related

Dual-arm Robot · ALOHA · Action Chunking with Transformers · Mobile ALOHA · Dexterous Manipulation · Mobile Manipulation

Sources

Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware (ALOHA / ACT)
Mobile ALOHA: Learning Bimanual Mobile Manipulation with Low-Cost Whole-Body Teleoperation

As of

2024-01