This robotic hand was built by Bruno Jau at JPL. I built a control system for the hand from scratch [1]. Because of severe budget limitations, I built a wire-wrapped multi-DSP system
to control the hand. We were able to get a closed loop servoing time of 2000HZ, with
a system having about 50 degrees of freedom.
This was a nice learning experience for me because
it was the first time I ever built a high performance multi-processor computer from
the ground up. I had to design all of the
interface circuits, the CPU modules, the scheduler, and the physical communication channels
(I used fiber-optic because of its simplicity and speed... and because JPL had a lot of it
around).
Here you see my hand in an exoskeleton glove master. The robotic hand is in the background.
The key innovation of the hand was supposed to be that the hand was the master and the exoskeleton glove was the slave. That is, if the finger on the hand was disturbed by a force,
this force would be reflected in the exoskeleton.
The robot hand could be commanded to move by exerting force on the exoskeleton.
I then mapped the force to a velocity command (this did not result in natural motion
of the hand due to various reasons (uneven calibration of the sensors etc) I introduced
a "regularization" factor into the velocity command that made the robots movement more
"human-like." This gave much better performance).
The video shows several seconds
of tele-operation.
As can be seen here, the hand was very human like. Each finger had 4 degrees of freedom.
The hand was basically built for grasping type tasks and some limited dexterous manipulation.
After building a system of this complexity it becomes absolutely clear
that automatic calibration and adaptation must be a part of
any complex machine.
For another picture of it and for related technology at NASA see:
