15-494 Cognitive Robotics
Spring 2007
Course Links:  Main Lectures Labs Homeworks

Tekkotsu Links:   Tutorial Resources Bugs CVS

Cognitive Robotics: Lab 6

Part I: Localization With the Particle Filter

In this part of the lab you will start with a pre-specified world map. You will use the MapBuilder to look around and build a local map, and then use the particle filter to localize the robot on the fixed world map.

Start with the file Lab6a.h. Arrange three plastic easter egg halves in an L-shape, with the two outboard easter eggs 100 mm from the central one. Make sure the colors of your eggs agree with the colors in the workd map set up by your code. Load the posture file STAND.POS so the dog is standing up; this increases its field of view and also makes the ground plane calculation more accurate.

Run the Lab6a behavior and observe the robot performing a visual search for the landmarks. Examine the local map to see what the MapBuilder found. Examine the world map to see what the particle filter concluded about the robot's position. Then move the dog to another location. Stop the behavior and start it again to process the new location. Hand in snapshots of your local maps and corresponding world maps. (You can use the "Save Image" button in the SketchGUI to save a picture of the local map or world map.)

Part II: Find and Knock Down A Soda Can

The Pilot can be used to search for an object by performing a map building operation. If the desired object isn't found, the Pilot can rotate the body and try again. Since the Pilot doesn't know what we're searching for, we must supply an exitTest() function to tell it when to stop searching. In this exercise we're going to look around for orange blobs, and then try to run over the biggest one.

Start with the file Lab6b.h. Use an upright orange soda can as your search target. Try starting the dog close to and facing the target, and see how well it can do at knocking it over. Then try starting the dog facing away from the target, and see if it can find it.

Part III: Gates of Heaven

Define a world map with two orange pillars (blobs) 35 cm apart, forming a "gate". Your task is to get the AIBO to walk through the gate without touching either pillar. To do this, you should define a Shape<PointData> on the world map that is a bit "outside" the gate. If you can get the AIBO to that point at the proper orientation, then you can use dead reckoning to proceed through the gate. You should first localize yourself on the world map, then use the Pilot's gotoShape request to travel to the gate entry point. Then use the Pilot's walk request to travel forward far enough to pass through the gate.

Note: gotoShape currently only navigates by dead reckoning and therefore doesn't work well enough to complete this exercise unless you're already very close to the shape you're trying to reach. So for now, focus your effort on localizing the robot and then printing out a message indicating the distance and direction from the robot to the gate entry point you defined; don't worry about executing the gotoShape request.

Dave Touretzky and Ethan Tira-Thompson