Finger Gaiting

Fingers release and reposition one at a time while the rest keep a firm grip, rotating an object through a large angle.

Finger gaiting is a form of in-hand manipulation: while a multi-fingered hand holds an object, one finger or a few fingers at a time break contact, move to a new position, and re-establish contact, while the rest of the fingers keep holding the object — much like legs taking turns stepping during walking. In-hand manipulation that never releases contact is limited by how far the finger joints can travel, so the object can only be rotated a small angle; finger gaiting continually breaks and re-forms contact instead, which in principle lets the object rotate through any angle. Model-based work in the 1990s by Leveroni and Salisbury, and by Han and Trinkle, among others, tackled this, but the instantaneous nature of making and breaking contact makes the model highly nonlinear and hard to apply on real hardware. More recent approaches mostly use reinforcement learning: OpenAI's Dactyl work learned finger gaiting, and Khandate and colleagues at Columbia University learned finger gaiting for precision fingertip grasps in 2021 using only proprioception and touch, with no external vision.

ExampleA dexterous hand pinches a cube between its fingertips and rotates it continuously around a vertical axis for several full turns: while three fingers hold it steady, the fourth lifts off, moves to a new position, and lands again; then another finger takes its turn, and the cycle repeats so the object keeps rotating.

Also called

Finger Gait, Grasp Gait

Related

In-hand Manipulation · Dexterous Manipulation · Dexterous Hand · Force Closure · Dactyl · Tactile Sensor

Sources

Khandate, Haas-Heger, Ciocarlie: On the Feasibility of Learning Finger-gaiting In-hand Manipulation with Intrinsic Sensing (ICRA 2022)