the whole biped if they didn’t weigh so much.
The supporting brackets for these servos were a bit
tricky to fabricate. It was necessary to keep the weight high
in the upper torso for an inverted pendulum effect. The
belts for the pulleys also needed to be aligned with the
existing hip pulley. You can see the two 1.5 inch aluminum
tubes used to hold the servos in Figure 1. With only 1/16
inch wide walls, these tubes are very light and very strong.
Supporting these tubes required replacing the four torso
vertical u-channels with one inch square tubes. Again, these
tubes are very light and very strong.
Once the upper torso was strengthened, it was time for
the arms. We wanted to simulate the movement generated
by swinging arms. The old legs proved perfect for this. They
were too weak as legs, but as arms they work well enough
even with the original servos. When the new servos arrive,
it should be easy enough to swap them out.
Only the knee servo mount needed to be modified to
allow the forearm to be attached. We couldn’t let the arms
have no hands, so as a joke we put a couple of gloves on
Watson. So, did this new hardware improve the design?
You bet. We charged up the batteries and ran a simple
program to stand and move the arms. This was mostly a
test of our battery capacity. Was there enough for the huge
power drain? Did our redesign of the legs create a more
stable platform? How long would the battery pack last?
We were pleasantly surprised that the robot ran for
about six minutes before collapsing. (Watch the video at
The next step will be the actual first step. We feel
confident that the new servos will have the additional
torque to get us walking. The DARPA challenge test that we
want to concentrate on will be Task 4: “Open Door, Enter
Building.” This will require a real hand. Full articulation is
not necessary since all three doors will have lever handles.
Biped Systems — Chris
These are probably the most important component.
Like many other bipeds, Watson uses 16 servos for primary
movement. There are a few additional servos used for the
pan and tilt of the cameras and head rotation. The arms
and legs are where we need the most power. It is
important that the servos match the torque requirements
for each joint.
Higher torque allows for better movement, but they tend
to weigh more. We are continually changing servos to get the
best balance of torque and weight for each joint. The servos
also consume the majority of the power. The servos often go
from idle to full power very quickly. This causes power spikes.
Isolating the servo power supply from the electronics prevents
voltage drops and possible unwanted resets.
Servo controllers come in a wide variety of choices. The
one we prefer is the standard Lynxmotion SSC- 32 controller.
The SSC- 32 has timing commands in addition to the simple
position commands. For example, the command “#5 P1600
T1000<cr>” will move servo 5 to position 1600 and take
one second to get there, regardless of how far the servo
has to travel. We have added a power and ground rail to
the back of the board to allow greater current for our high
power servos. We found it is sometimes an advantage to
use more than one servo controller. This way, we can split
up the power required and use a smaller group of servos.
The SSC- 32 uses a serial interface, making it easy to
interface to our CPU.
The biped’s electronics are fairly straightforward and
use a collection of mostly off-the-shelf parts.
We have two custom boards to use for our load cells
and the auto-balance feature of each foot. We also have a
hand-built power board that manages important things like
a safety power-off switch and fuses.
The CPU is the biped’s brains. We have used everything
from Arduinos and Raspberry Pis, to ODROIDs. Each has
advantages and drawbacks, but for the most part it really
doesn’t matter as long as you can control all of your
electronics and do all of your processing in real time.
The Arduino was fine to start out, with its I2C bus and
multiple serial ports. When we needed to connect a
camera, we had to move over to the Raspberry Pi. This also
has both an I2C bus and serial ports, and in addition has
USB ports and a special camera connector.
When we needed more processing power to process
stereo cameras for mapping and navigation, we moved up
to the ODROID. All of our code is in C, so it recompiles
easily and runs on all of the above platforms.
The IMUs — or Inertial Measurement Units — give
SERVO 11.2013 39