We took the i0800 servos that didn’t have the high
torque and made feet that can turn. This enables us to
turn the whole body as we walk. We can also bring the
foot completely forward from behind without the toes
clipping on the opposite leg. The frame sequencer that
drives the biped simple turns the long toes backwards
during that move so the foot is narrower during that
move, and spins them sideways again before the foot
The leg movement is driven by three servos that
simulate and ankle, a knee, and a hip that can only move
forward and backwards. Unfortunately, after the brackets
arrived, I noticed that they were only designed for 90
degree movement and we needed 180 degrees. In
addition, since they were not designed for this
application there didn’t seem to be any way to stack-mount them end-to-end to build a human style leg.
Removing a small piece from the main base of the
bracket solved problem one and we found we could get
the 180 degrees of movement. Drilling new holes on the
back plate and mounting a one inch square aluminum
tube to hold two back plates head to toe solved the
second problem. (See Figure 2.)
We now had a workable platform. The upper torso
was not able to support these new servos or the newly
designed arms that now had hands built by Glenn and
Ansis. So we rebuilt that too. (See Figure 3.) Whew!
It was also nice to add a camera with its two-axis pan
and tilt to start testing our object-tracking software. Chris
is happy to finally have a working test rig for that.
We were all very excited — too excited. We were in a
hurry and we rushed the testing. Everything was going so
well that we started taking shortcuts. Even though we
had WATSON tethered and floating in the air, we made a
crucial error and ran the wrong untested sequence. We
normally only ran with three people. One person would
run the control program; one person would catch the
biped if it fell; and one person would be on the E-stop
switch. This time, we only had two people.
When we ran the wrong sequence, the servos
lurched, the left toe caught the right leg, and the servos
just wouldn’t quit. We were not fast enough to hit the E-stop and a horrible grinding sound came from one of the
gearboxes. Of course, it was the Friday evening before
the qualifying videos were due on the following Thursday.
With less than a week and without a spare, we were
dead in the water for at least three days.
Monday came and I had new parts shipped
overnight. Tuesday was mostly spent on rebuilding, and
we starting testing the walk that night. There was still
two days left.
The First Step
I have been working on this project for seven years,
and have gone through at least six full rebuilds. Even
though we knew deep down inside we would not meet
the deadline, we did not quit. We worked on a static
model walk and by Thursday afternoon — about an hour
before the deadline — WATSON took its first step. This is
the first platform out of six that could successfully repeat
a step without falling over.
50 SERVO 12.2013