Willow Garage announced it has taught its two-armed, mobile
PR2 robot to repeat activities such as pouring water in a glass even if the
objects involved are of different sizes (watch video).
That’s second nature for humans, but robots learn by being
put through a series of motions and then repeating them precisely—which doesn’t
work if the size or distance between the objects has changed. A typical robot
is more likely to just pour water on the floor rather than correct for a change
in the distance between pitcher and glass.
The key to the ‘bot’s new flexibility is changes to its
operating system that allow it to generalize movements and then apply the basic
movements it has learned to a situation that’s similar but slightly different—such
as when someone has moved the glass or the “water” is in a soda can instead of
a pitcher.
PR2 has also demonstrated an ability to find and
plug into electric sockets while navigating unaided through eight doors,
obstacles in a crowded office environment and humans who suddenly stood in its
way.
Willow Garage, an open-source robotics developer, is offering
10 PR2 ‘bots to developers who want to build significantly on the robot's existing
abilities.
Researchers at the Max
Planck Institute for Dynamics and Self-Organization have announced they’ve
harnessed the power of chaos itself to—well, to change the gait with which a
robot walks (watch video).
That’s far more difficult than it seems, however.
Robots, like humans, learn different gaits for different
challenges (walking up a slope, for example, or down a slope or around obstacles).
In humans, those gaits—each a separate set of behaviors learned once and then
put back into practice when the situation warrants—are controlled by neural
circuits called "central pattern generators" (CPG).
Roboticists have replicated that approach, but have
generally needed a separate electronic CPG
subsystem for each gait.
The Max Planck researchers created a CPG
that, in its natural state, produces a chaotic activity pattern. The pattern
changes according to input from the robot’s sensors, and can shift from one to
another of several predefined patterns until it finds the one that requires the
least energy. In other words, it receives one set of input while walking across
a level floor, and the chaotic activity of the CPG
is quelled by a stable level of energy output. When it reaches a slope, energy
output goes up, and the CPG gets a little
chaotic and switches around until it finds the right behavior for the
situation.
The next step, researchers say, is to build in memory so the
robot can remember what it was doing even after the initial sensory input is
gone. Without that memory, it can’t walk over a large obstacle because it would
essentially forget what it was doing—actually, what obstacle it has to overcome
and the behavior pattern required—once it had climbed partway over and lost sight
of the obstacle on which it was now perched.
Despite those advances, Korean developers may be ahead of
Willow Garage and Max Planck.
After two years of work, researchers at the Korea Institute
of Science and Technology say they’ve created a walking robot maid
that’s able to recognize humans, recognize chores that need to be done, clean a
home, put clothes in a washing machine and heat food in a microwave.
Which is all very impressive, except for the big pink apron
and white skirt they put on it, which kind of detract from the grandeur of it
all.
