RAFS - Progress
Progress
CS 425
Here is a some links from our prototype in CS 425. It consisted of getting
the robot to move in a pre-defined square. We had some technical problems with
the robot, which kept us from testing it to our full capabilities. The robot
was shipped back about a week before our prototype was due. Therefore we had
to get our prototype done ahead of time. We spent a lot of time in one weekend
to get a majority of the code written. A video of our prototype and the code
behind it follows:
CS 499
September 15
Here is a little sample of our object (chair) recognition so far. It includes
a crude video capture from our robot development environment, along with a video
from a regular camera. These two videos were recording at the same time, which
gives the opportunity to see what was happening from two different angles. On
the video from the camera mounted on the robot, we have the environment in which
our project will be developed open, which is Saphira. The black and blue dots
are where the sonar sees an object, like the wall to the right of the robot.
The heading of the robot can be determined by the little red rectangle, which
is the front. The actual code for this module is in the black window. The camera
shows blue highlighted pixels, which represent "recognized" colors.
The color we are using is pink because it is easily distinguishable from other
colors in the room we are actually working. In this example, the chair is placed
within the camera's view. The robot begins by repositioning the camera and begins
to persue the chair. It will then begin to turn toward the chair as it is approaching
it, continuously adjusting to put the chair directly in front. When it gets
within a reasonable distance from the chair, it stops.
September 24
Here is the final version of our Object Recognition module, which is now integrated
with the wander module. It will wander around until it sees a chair, then it
will approach the chair. When the chair is within a certain distance from the
robot, the gripper will open and close again to let you know it found a chair.
The chair will also be logged, along with the coordinates of the chair. Localization
plays a big part in how accurate the logging system is. We will get it to be
more accurate in Release 2.
September 26
Here is a copy of our Chair Movement module. It starts at a point with the
chair already grasped. It then progresses forward until an object gets in front
of the chair. Once this object is detected, it turns a random direction to go
around the object. At this point, we are assuming that the coast is clear on
both sides of the object. It turns 90 degrees, then goes around the object.
Once it is around the object, it goes until a certain distance is covered (passed
as an argument to the function call). Once that distance is travelled, the robot
turns 90 degrees back, and moves forward until he is along the same path (roughly),
and he turns to face the same direction as he was when he started. Localization
will make this module more accurate in stopping with respect to the same line
as it started.
October 3
This module began with the Object Recognition module from above. It took that
module one step further and actually grasps the chair. The video skips the repeated
part from above, and begins when the robot is aleady in front of the chair.
A major assumption that was made for this module since it is the research phase
of gripping a chair was that one of the legs of the chair must be directly in
front of the chair from the robot's view. This eliminates the problem of realigning
the robot to approach the chair from the correct angle. A few ideas for algorithms
have been conjured at this stage. One of these algorithms will be implemented
in Release 3.
October 14
We captured a small segment of this module running on video. We kept it in
its native resolution and format to preserve quality. Since this format is large,
we kept the video short. It is important to see the details of the output window
to realize what is going on. Also, conversion to mpeg would reduce the quality
of the video to the point where laser readings would be hard to decipher.
This is a sample run of the robot selecting random points in EB2029 to "wander"
to. In the background the object recognition module is looking for chairs. In
this test run the robot does not see a chair until the near end of the video.
When the object recognition modules see a chair, it interrupts the wander module
and moves the robot in the direction of the chair. This example employs the
new module interaction scheme we have devised.
October 15
In this module, we need to put a chair, once gripped, into the correct place
by the desk. The goal of this part is to fine tune a point to use for a desk.
We start with the chair gripped, then move toward the point, with the assumption
that nothing is in the path. Once we get to the desk, we open the gripper to
let you know it is in the correct place.
November 5
In this module, we start out with the chair in sight and within the given distance
the object recognition module leaves off. The robot will then have two cases:
the leg is lined up, or it is not lined up. If it is lined up, it simply grabs
the chair. If it is not lined up, it turns 90 degrees to the right, and goes
in a circle around the chair, until the leg is lined up with the center of the
chair. It then turns back toward the chair and grabs the chair. It continuously
realigns itself with the center of the chair, so it does not miss the leg.
