Page 65 - Modern Robotics Building Versatile Macines

LEARNING TO WALK 45

dog—or perhaps better, “mule”—will be capable of intelligently
finding its way around the terrain as it follows troops in combat.
Able to carry supplies and ammunition, such robots could extend
the range and duration of combat patrols in areas where vehicles
are either not available or unable to clamber over terrain, such as
the mountains of Afghanistan. (In addition to Boston Dynamics,
the BigDog development team includes MIT, Harvard, and
Stanford.)

be used to generate animation sequences. This worked by creating a
“virtual robot” and subjecting it to real-world physics while simulating
real-world mechanics. In a 1991 paper for the ACM Computer Graphics
journal, Raibert and IBM researcher Jessica Hodgins explained how
they constructed animation models for bipeds (running and hopping),
quadrupeds (trotting, bounding, and galloping), and even kangaroo-
style hopping. As with the original robot models, the general approach
was to simplify the system by focusing on a single moving leg or
coupled pair of legs. A leg is modeled as rigid segments connected by
joints. Much of the complexity arises from dealing with the way energy
is stored and released by muscles (actuators) and transformed via the
joints, as well as accounting for the “springlike” characteristics of the
feet as they contact and release from the ground.
Since that original work, great advances in computer power and
the availability of many other sophisticated animation models (such as
for human facial expressions) have combined to enable products from
Boston Dynamics and other companies to include highly realistic por-
trayals of the human body under many different types of conditions.
The connections between legged robot locomotion, motion simu-
lation, and animation are a striking illustration of how traditionally
distinct fields such as biology, mechanical engineering, and computer
graphics can develop together by tapping into a growing body of
physical understanding. The results are likely to include more useful
and versatile robots as well as educational simulations and games with
greater accuracy and realism.