Modeling and Animating Human Figures - Computer Animation: Algorithms and Techniques - page 295

then the upper arm/shoulder. Adding motion of the torso including the clavicle and spine, which
involves more DOFs, makes this task simpler to accomplish, but also makes the specification of real-
istic joint angles more complex. It is difficult to determine exactly what rotation should be assigned to
which joints at what time in order to realistically model this motion.
Interaction between body parts is a concern beyond the determination of which joints to use in a
particular motion. While viewing the arm and hand as a separate independent system simplifies the
control strategy, its relation to the rest of the body must be taken into account for a more robust treat-
ment of reaching. Repositioning, twisting, and bending of the torso, reactive motions by the other arm,
and even counterbalancing by the legs are often part of movements that only appear to belong to a
single arm. It is nearly impossible for a person reaching for an object to keep the rest of the body
in a fixed position. Rather than extend joints to the edges of their limits and induce stress, other body
parts may cooperate to relieve muscle strain or maintain equilibrium.
Arm manipulation is used in many different full-body movements. Even walking, which is often
modeled as an activity of the legs only, involves the torso, the arms, and even the head. The arm often
seems like a simple and rewarding place to begin modeling human figure animation, but it is difficult to
keep even this task at a simple level.
9.2.5 Reaching around obstacles
To further complicate the specification and control of reaching motion, there may be obstacles in the
environment that must be avoided. Of course, it is not enough to merely plan a collision-free path for
the end effector. The entire limb sweeps out a volume of space during reach that must be completely
devoid of other objects to avoid collisions. For sparse environments, simple reasoning strategies can be
used to determine the best way to avoid obstacles.
As more obstacles populate the environment, more complex search strategies might be employed to
determine the path. Various path-planning strategies have been proposed. For example, given an envi-
ronment with obstacles, an artificial potential field can be constructed as a function of the local geom-
etry. Obstacles impart a high potential field that attenuates based on distance. Similarly, the goal
position imparts a low potential into the field. The gradient of the field suggests a direction of travel
for the end effector and directs the entire linkage away from collisions ( Figure 9.13 ) . Such approaches
are susceptible to local minima traps that need to be dealt with. Genetic algorithms, for example, have
been used to search the space for a global minimum [ 48 ]. The genetic fitness function can be tailored to
find an optimal path in terms of one of several criteria such as shortest end-effector distance traveled,
minimum torque, and minimum angular acceleration.
Such optimizations, however, produce paths that would not necessarily be considered humanlike.
Optimized paths will typically come as close as possible to critical objects in the path in order to
minimize the fitness function. Humans seldom generate such paths in their reaching motions. The
complexity of human motion is further complicated by the effect of vision on obstacle avoidance.
If the figure “knows” there is an object to avoid but is not looking directly at it, then the reaching
motion will incorporate more leeway in the path than it would if the obstacle were directly in the
field of view. Furthermore, the cost of collision can influence the resulting path: it costs more to
collide with a barbed-wire fence than a towel, and the path around these obstacles should probably
be quite different.