The robot moves in its enclosure, rendering what it sees into the virtual
world. It turns as it sees the
walls, then sees a chair through the open door.
What ADAPT constructs in its 3D world model as it
executes the movement in the previous clip. It builds the
virtual walls correponding to the real walls it sees, and
the door it perceives, then
recognizes a chair and adds it to the virtual world.
A ball is rolled between two parallel boards. In the initial phase, the ball's
position and velocity are captured. Then the virtual world is shown and
the ball is rendered and its trajectory is predicted. Finally the rest of
the ball's trajectory in the real world is shown then compared with the
virtual world. The real and virtual video are slower than real time for
A ball is bounced off two angled boards and then between two parallel boards.
In the last portion, the virtual world is run a little faster than real time
to show the predicted path.
The ball is bounced off the angled boards a few more times. This is closer to
of the RS component controlling a robot arm and
camera to pick up an erratically moving small black block. The block is
being manipulated by a human using a magnet under the table. The
dynamic reactive abilities of RS are integrated with Soar's problem
solving and learning abilities to form the core of ADAPT.