Homing

The environment is a square with sides measuring 420 meters which contains six types of obstacles. A total of 301 obstacles are randomly place d in the environment. At any given time there is one goal, and when all flock members reach it, a new goal is randomly generated; this process continues until eight goals have been generated and reached. The experiment involves 40 flock members, which are initially placed according to a Gaussia n distribution around the center of the square environment.
The simulation is updated every 100 msec. The roadmap is built using the MAPRM method to generate 400 roadmap nodes and we attempt to connect each node to its 4 nearest neighbors. Also, a simple 3D case is tested with 40 flock members.

In the movies, the roadmap is displayed with the lines. The first movie shows the roadmap approach in a complex 2D environment. The goal location is represen ted with a red line. The second movie shows the roadmap approach in a 3D environ ment.

Movie:

Covering

In this experiment, we compare basic flocking behavior, roadmap-based behavior, and an ideal variant of the roadmap-based behavior which has dynamic knowledge of the undiscovered regions.
The environment (80*100) is populated with 16 obstacles (6 types of obstacles) and in total 24% of the environment is occupied by obstacles. 50 flock members are simulated and states are update d every 100 msec. A bitmap is built to record discovered/undiscovered information. A bitmap cell is discovered when it is inside the sensory range of any flock member. We set the radius of the sensory circle as 5m. For the roadmap, 120 nodes are sampled and connections are attempted to each node's 4 nearest neighbors.

In the movie, the flock starts at the center of the environment. The road map is represented in the white lines. Initially all the environment is dark. As the members visit new regions, the covered areas are highlighted. Circles ar ound the members represent their sensory range.
The first part of movie shows basic behavior. It can be seen that some parts of the environment is not covered since dark regions around the buildings are left on the top left corner at the end of the movie. However when a roadmap based app roach used, all the environment is covered as shown in the second part.

Movie: Covering (mpeg 10.9MB)

Goal Searching

In this experiment, the roadmap-based behavior is compared with simple flocking behavior that has only local information about the environment and no knowledge of the goal position and with an ideal variant of the roadmap-based behavior that has a priori knowledge of the position of the goal. The environmental is the same as that used in the covering experiment.

The first part of the movie shows the basic behavior. The goal (which is unknown ) is represented by a yellow line. The flock couldn't reach the goal. The second part shows the roadmap approach where the goal location was known. As it can be seen, all the members converge to the goal location immediately. In the third p art, the goal location is unknown and individual members update the roadmap edge s' weight if they find the goal. At the beginning the members are searching the space. As individual members reach the goal they increase the weight of edges th at reaches the goal. At the end all the members converge to the goal.

Movie: 2D Goal Searching. (mpeg 10.6MB)

In addition, a group of snake-like deformable objects are simulated in a 3D environment. Flock members need to pass through two ho les to find the goal. 20 deformable objects are simulated in a 650*450*300 bounding box.

Movie: Deformable Object Goal Searching. (mpeg 9 .4MB)

Shepherding (see also Specialized Techniques for Shepherding Behaviors)

In this experiment we used the same environment as the homing experiment. In this experiment, an outside agent (dog), steers a herd of sheep. The sheep show basic flocking behavior (they tend to stay together) while trying to avoid the dog.

The movie shows a simple environment. The roadmap is represented in the white lines. The goal is to move the herd to the stable. The radius of individual gro ups are represented by a circle. If some members gets far away from the herd, th e dog steers them back to the herd. The dog uses roadmap to find a path for the herd to the stable. It steers the herd towards subgoals placed in the path. Thes e subgoals are represented with a red sphere in the movie. The steering location for the dog for either moving the herd towards to subgoal or moving the separat ed members towards to the herd is represented with a light blue sphere.

Movie: Shepherding (mpeg 10.8MB)

Narrow Passage


This experiment shows that flock's behavior depends on the surrounding environment. Different group formations may be used in relatively open areas than when passing through narrow regions.

A naive way to achieve narrow passage traversal by the flock is to use the homing behavior. One drawback of this approach is that flock members may bunch up and conflict with each other as they try to move through the passage. Using rule based roadmap, we first assemble the flock in front of the narrow passage, and then select the closest flock member to the entrance to the narrow passage as the leader. Then, the remaining flock members are arranged into a queue that follows the leader.

Movie: