|
Building synthetic brains capable of human level
discovery and invention...
|
Robots That Learn from Scratch
Summary - Creativity
Machines built from STANNOs form the basis of robotic control systems that may
rapidly learn all of their behaviors from scratch. In other words, our robots do
not require that computer programs be written by human beings. Given very broad
goals, such as the objective of moving forward, the robot learns to walk from
the lessons of its own self-generated experiments. The autonomous development of
other robotic skills proceed in the same general way, such as in finding a
pre-specified target within the robot's environment. Essentially, the robot
experiments with a number of self-conceived approaches to the problem,
cumulatively learning from its successes and failures to attain its goals.
Details - In the two
movies shown here, the first (left/top) is of an 18 degree of freedom
hexapod crawler controlled by a STANNO-based Creativity Machine after just 1
minute of cumulative training. The robot utilizes its onboard sonar to calculate
its progress toward a wall located at the top of the frame. Note that the
robot's body has been deliberately constrained to move in one dimension to avoid
misleading sonar returns. The tether trailing behind the robot and containing
the serial cable to a PC, has been festooned to avoid torques from it. At this
stage of training, after 1 minute of its self-experimentation, the robot slowly
and tentatively flails about, making very little forward progress.

In the right/bottom movie, the
robot has been experimenting and learning from its successes and failures for 8
minutes. In contrast to its walking skill after 1 minute, it has begun to
develop efficient forward walking skills. Using similar procedures, the robot is
able to self-invent backward, left turn, right
turn, left shuffle, and right
shuffle strategies.
For more IEI robotics activities
see:
|
©
2009,
Imagination Engines, Inc. |