MATH & TECHNOLOGY
Avatars watch YouTube to learn moves
Computer program improves on motion capture technology
BY MARIA TEMMING
Animated characters can learn from
online tutorials, too. A new computer
program teaches avatars new skills
such as dances and martial art moves
from You Tube videos. This kind of system, described in the November ACM
Transactions on Graphics, could render
more physically coordinated characters
for movies and video games, or serve as a
virtual training ground for robots.
“I was really impressed,” says Daniel
Holden, a machine-learning researcher
at Ubisoft La Forge in Montreal. Render-
ing accurate, natural-looking movements
based on everyday video clips “has always
been a goal … in this field.”
Animated characters typically have
learned full-body motions by studying
motion capture data from a camera that
tracks markers attached to actors’ bodies.
But this method needs special equipment
and often works only indoors.
The new program leverages a computer code known as an artificial neural
network, which mimics how the human
brain processes information. Trained on
about 100,000 images of various poses,
the program first estimates an actor’s
pose in each frame of a video clip. Then
it teaches a virtual avatar to re-create
the actor’s motion using reinforcement
learning, giving the character a “reward”
when it matches the actor’s pose.
Computer scientist Jason Peng and
colleagues at the University of California,
Berkeley fed You Tube videos into the
system to teach characters to do somersaults, backflips and other stunts.
Even characters with bodies drastically different from those of their human
video teachers mastered the motions.
Characters also performed under conditions not seen in the training video, like
moving across terrain riddled with holes.
The work is a step “toward making
In a new computer program, virtual characters
learned full-body motions, such as cartwheels,
by watching people in video clips.
motion capture easier, cheaper and more
accessible,” Holden says. Videos could be
used to render virtual versions of outdoor
activities or to create lifelike avatars of
large animals that would be difficult to
attach motion capture markers to.
These animated characters still struggle with nimble dance steps such as
the “Gangnam Style” jig and can learn
from short clips featuring only a single person. Computer scientist David
Jacobs of the University of Maryland in
College Park looks forward to future virtual avatars that can reenact longer, more
complex actions, such as pairs of people
dancing. “This is only the beginning,”
Jacobs says. s
See how virtual characters learn new moves at bit.ly/SN_Avatar
Neutrinos reveal Earth’s mass
Puny particles have given scientists a
glimpse inside Earth. And for the first
time, physicists have measured the
planet’s mass using neutrinos.
Scientists have previously studied
Earth’s innards by quantifying the planet’s gravitational pull and by studying
seismic waves that penetrate the globe.
Neutrinos provide an independent test.
With data from the IceCube neutrino
detector at the South Pole, researchers
estimate Earth’s mass is about 6 trillion
trillion kilograms. That’s in agreement
with traditional measurements, a trio of
physicists reports online November 5 in
The team studied high-energy neutrinos. These neutrinos can zip clean
through Earth, but sometimes smash into
atomic nuclei and get absorbed instead.
NEWS IN BRIEF How often neutrinos are stopped in their
tracks reveals the density of what they’re
traveling through. Neutrinos that arrive
at IceCube from different angles probe
different layers of Earth. By measuring
how many neutrinos came from various
angles, the team inferred the densities of
different parts of Earth and its mass.
The technique might one day show
whether any of Earth’s mass comes from
dark matter, a shadowy substance that
scientists believe must exist to account
for missing mass observed in measurements of other galaxies. — Emily Conover
GENES & CELLS
Calorie burning tied to time of day
People at rest burn about 129 more
calories in the afternoon and evening
than in the early morning, scientists
report in the Nov. 19 Current Biology. But
morning is better for burning carbohy-
drates, whereas fats are more likely to
be burned in the evening. These findings
add to evidence that when people eat
and sleep may be as important as what
they eat (SN: 10/31/15, p. 10).
The study followed seven people kept
in windowless rooms for three weeks.
Each night, the seven went to bed four
hours later than the previous night. The
schedule change let the researchers
study the natural body rhythms of each
Clear rhythms emerged for when
people burned calories, with some variability. Resting calorie burning peaked on
average around 5 p.m., with some people
peaking around 2 p.m. and some at 8 p.m.
The lowest calorie burning came around
5 a.m., but ranged from about 2 a.m.
to 8 a.m, neuroscientist Jeanne Duffy
of Brigham and Women’s Hospital in
Boston and colleagues found.
Irregular schedules can interrupt these
rhythms, which can throw off metabolism,
causing people to burn fewer calories and
gain weight. — Tina Hesman Saey