Robot Figures:  Design Evolution
To bend happily, tubes and tendons need to get parallel, like your 4 fingers.  And if I wanted the robot to be able to run for hundreds or thousands of hours without an overhaul, a radically different system was needed.  The stiffened tubes had to go.  Accepting such a "back-up and change direction" is hard, but the failed technique wasn't a total waste, it taught some important lessons and helped a lot with visualization of the next design. Also, the stiffened tube technology could be used well in certain other applications.

The new direction has similarities to bicycle cabling - the tendons are mostly exposed, but go into in conduits to traverse turns.  A system of parallel, axially stiffened conduits built as a module was developed.  This was called a [link]"bender", and it was immediately obvious that this was a good way to go.  The bender could flex thousands of times without deterioration, and if fine cable (instead of single strand hard wire) was used as the tendon, we could have a very durable system.  The first test rig for the bender concept (right 2 photos)  is very close to the final bender design.
Here, presented in some detail, is a case study of how a design evolved. (Note on photos: these test rigs were photographed long after their use, many are incomplete)

People are modifiers, not creators.  Complex forms and systems cannot be conceived entirely in the imagination, but when we see and feel something, improvements to it are readily envisioned.  As several generations of prototypes are constructed and improved upon, a design will progress from naive to well-adapted.

The Slave Zero robot started out as a vague idea for hobby servo based small figures which would perform in dioramas.  The motions of the servos would be transmitted to the figure's joints by fine hard wires inside teflon tubes.  These tubes would be bundled and go where they needed to go.

Locating the servos off-limb was, from the start, considered essential for aesthetic and practical reasons; I wanted a lithe,  magically moving figure, not a squat, opaque pile of servos.  So the project became mainly a motion transmission challenge.
This rig was the first try at sending motion from a servo to somewhere.  The servo (not present) would drive the pivoting arm on the right, to the actuated joint on the left via a hard fine steel wire inside a teflon tube.  (The tube path looops up out of the picture).  The tube ends were gripped and the wire inside was free to slide as the servo pushed and pulled it.
The idea above works for very simple, lightweight systems, but a multijoint figure is not so light or simple.  There are a few key problems.  The first is that a simple lever from the servo to the wire only allows a small part of the servo's 180 degree range to be used. A driver DISK instead of a lever would address this issue. The second issue is that wire pulls better than pushes.  Thus a 2 TUBE, pull-pull solution with a driver disk which the wire attaches to. There was still hope at this early stage that the driven end of the system could stay simple - no driven disk.  This system did not work smoothly.  
A 2 disk solution (a driver disk at the servo and a driven disk at the joint) was first tried in this rig.  Also, counterbalancing was explored which could ideally reduce joint forces to near zero.  And, this was the first attempt at a 2 axis system - a sort of shoulder and elbow. 

The actual pivot design is very simple:  aluminum structure rides directly on stainless steel shoulder bolts.  Counterbores in the aluminum were made such that the shoulder bolt had a tiny amount of clearance - the joint pivoted freely but didn't wobble too much.

For light, slow moving systems, such aluminum on steel "bearings" can work.
The counterbalance concept above was obviously too bulky and unworkable. A different approach to reducing joint force was tried out: going "ultralight" with thin wall tubing based structures. The system worked pretty smoothly, but clearly could not handle much more weight. And more weight would certainly be needed with an anthropomophic robot with wrists and hands.

The teflon tubing ends are held in place with epoxy, and the tendon ends are held to the driven disk with a pair of set screws.
There was a key problem with the "hard wire in teflon tube" strategy for transmitting motion - while the wire is in tension, the tube is in compression and will buckle. The Teflon tubes simply cannot withstand compressive force.  The diagram at left illustrates the problem. 

The Teflon tubes needed some form of axial stiffening - a "backbone" - if this system was going to work.  Bicycle cable conduits are axially stiff, but way too unflexible. Good flexibility was needed since a bundle of these stiffened tubes would have to  traverse joints, some bending over 120 degrees. 

The first attempt at stiffening was 2 tubes and 1 wire, with wrapping to hold it all together.  It didn't wear well, and was not stiff enough.

Small, flexible stiffened tube was needed, and a machine was built to help create such stuff. 
With the new flexible incompressible tube from the wrapping machine  available, it was time to make something more complex.  This rig had a 3 degree of freedom shoulder and an elbow, and it was hooked to a computerized servo-control system. It actually looked like an arm and it worked. This rig was very lightly built, and flimsy. And most troubling was that the lay of the tubes was going to lead to quick wear-out due to harsh flexing and rubbing. It seemed that to minimize the bending and rubbing stresses on the tube bundles, they must go through the CENTER of the joints.
This next arm iteration had the tube bundles routed through joint centers to minimize bundle thrashing, and was a little more substantial in build strength.  Nested brass tubes formed the revolute joints as well as being "bones".  Aluminum pulleys were pressed and/or epoxied to the brass tubes.
The next arm (above, and compared to predecessor at left) was much sturdier and included ball bearings on several axes.  This was pretty much the best that could be done with wire-in-stiffened-tube technology. It worked well, but clearly would not work long.  These tube bundles just didn't like repeated bending - bundles never do.
The above rig solved the bending issue nicely. However, there was a serious cable friction issue.  The cable tendons were informally layed in, and rounded corners over Teflon cylinders. This is OK for a few cables, but when there are several, they tend to crowd each other and get "intertwisted".  To reduce friction to a minimum, each cable needed to have its own path, and be prevented from touching its neighbors.  Thus, the "revector block" was born.  Revector blocks (red objects in photo below) have Teflon tubes embedded in them which guide each cable just where it needs to go. Revector blocks plus benders allow cable runs to be optimal: parallel for bending (through hinge joints), and tight hexagonal grouping for twisting (through revolute joints).
At this point, with parallel benders, revector blocks and 7 x 7 cable tendons, the technology was reasonably mature. The Slave Zero robot (right) was constructed.  A year later, assorted slight improvements were made to the design and Slave One was built.  These 2 machines have operated reliably since 1998.
UP to Robotic Figure Overview
NEXT Robotic Figure Topic
works of  Carl C Pisaturo