Hand controller

Requirements

Simple blind grasps (essentially what we had up to now)

1. Reduced backlash (removing springs in hand)
2. All three fingers simultaneously meet
   • in a point
   • on the (virtual) circumference of a ball
3. Current control is used to excert a given grasp force

Advanced blind grasps

1. Proximity control: the proximity sensors are used to have all 3 fingers maintain the same distance to the object.
2. This distance can be continuously reduced so that eventually all fingers have object contact simultaneously
3. Impedance control: the finger torque sensors are used to maintain contact with a grasped object

Learned grasps

Using known grasp points on the object or manifold matching etc.

1. Fingers can be placed on known points on the object, i.e. they can be moved to given cartesian coordinates
2. The middle finger (thumb) is placed first, using its 1 DOF and the 6DOF of the hand pose
3. The other two fingers (left and right) are moved using the 2 DOFs of the finger plus respective palm Of course the fingers can not be moved to any 3D point, just to a point on a plane through the finger.

Skill learning

This replicates the skill learning and execution on the Kuka arm plus Schunk hand

1. Two fingers (left and right) can move along a trajectory using the associated palm DOF, while the thumb (middle finger) plays a supporting role)