Prepping the four HUBO robots for
the photo shoot.
The new HUBO is enabled with much more human-like
motor skills including faster, more realistic arm movement
and stretching legs that mimic human bi-pedal movement
more closely. Natural human walking requires less energy
than typical robot walking, so there is an energy savings.
This is calculated using the Zero Moment Point trajectory
calculation. The HUBO robot now has an increased walking
speed of 1.4 km per hour and can now run at up to 3. 6 km
The new HUBO platform leverages the following
technologies: a smart power distributor and lithium polymer
battery that is +48V, 8A; a shape adaptive hand that is
tendon-driven with five fingers (one degree of freedom per
finger) and an F/T sensor at the three degree of freedom
wrist; an IMU (two axis) with Kalman filtering and a rate
gyro and accelerometer; a two channel BLDC controller that
is 90 mm x 65 mm with a 16-bit microprocessor; a CAN
interface; an A/D converter; over-current protection;
automatic return to initial position; a BLDC motor amplifier
that is 90 mm x 90 mm 200W and 48V; and a full bridge
Drexel received one HUBO initially and six HUBO 2s
later on. The last six are identical but they each have
individual quirks. One has an ankle motor that is a little off;
another has a different computer inside, but the platform is
otherwise the same, says Dr. Kim. Drexel has given each
one a number to tell them apart.
The first of the two HUBOs (introduced in 2004) did
not have the power efficiency, custom hardware, and circuit
boards available on the newer HUBO 2. The HUBO 2
operates on battery power for nearly an hour. KAIST has
lowered the center of gravity on this robot so that it will
not fall over so easily. HUBO 2 has increased motor
efficiency and faster processors in its internal computer.
One of those processors is a standard PC 104 platform
base. There is room for two processors in the robot: one for
command, control, and movement; and the other for
sensing, cognition, and higher-level tasks. “The challenge is
in having the two processors communicate with each
other,” comments Dr. Kim.
Dr. Kim has developed algorithms for tracking the beat
in music based on the audio. This is a hard problem to
solve. It is difficult to teach a computerized robot to do that
in a robust way. “We are working on having the HUBO
robots play simple percussion instruments that we have
built out of PVC pipe. We are working on having them play
the notes and listen to the notes to see whether a note was
a good note. This involves audio and force feedback
capabilities,” explains Dr. Kim.
Dr. Kim is using machine learning and as HUBO makes
lots of differing strikes on the instruments, Dr. Kim presents
some to the robot and tells it that these were good or bad
notes. The force feedback sensing from the robot’s hand
and wrist together with the sound of the note tell the
characteristics of the different notes.
SERVO 10.2012 11