Robot Software Design and Results

Figure 6-14 shows the ALLIANCE-based control of the robots under the bounding overwatch mission. At the beginning of the mission, the only behavior set whose sensory feedback is satised is the join-group behavior set. Since this task must be performed by all robot team members, the motivational behaviors in all the robots ac-tivate this behavior set at the beginning of the mission This behavior set is important because it allows the team of robots to divide themselves into two equal subgroups.

This division is accomplished using the following simple rule in each robot:

Wait a random length of time t (between 0 and some prespecified maximum time).

While waiting, monitor the messages of robot team members, keeping track of the number and type of robots in each subgroup so far.

After the random wait period is over, do the following:

1. Select the subgroup with the minimum number of my type so far.

2. If the two subgroups have equal distributions of my type, Then: Select the group with the fewest members,

breaking ties arbitrarily.

3 Broadcast the group I have joined and my robot type.

The prespecied maximum time of a wait should be long enough to reduce the like-lihood of interference between robot messages to an acceptable level. Once a robot has joined a group, it moves to the prespecied gathering location for its group. The rst snapshot of gure 6-15 shows the initial location of a group of eight robots | four of each of two types. The second snapshot shows the robots dynamically divid-ing into the two groups and movdivid-ing to the specied gatherdivid-ing locations (indicated in gure 6-15 by the two small triangles closest to the robot starting locations).

6.3. THE BOUNDINGOVERWATCHMISSION 175

Figure 6-14: The ALLIANCE-based control for the bounding overwatch mission.

The preconditions for the emerge-leader behavior set to activate in robot ri are that (1) ri has arrived at (more accurately, visited) its group's initial gathering point, and (2)ri's group does not yet have a group leader. If these conditions are met, then the emerge-leaderbehavior set is activated. The result is that the rst robot to arrive at its group's gathering point becomes that group's leader. An interesting side-eect of this denition is that if, at any point in the future, robotri's group loses its leader, thenri will become motivated to emerge as the team's leader. Since many other team members will also have this motivation, the relative rates of impatience across robots will determine which robot actually does emerge as the leader. Ideally, the rates of impatience are set such that the robots which make better leaders become motivated more quickly to become a leader. If there is ever a tie in which more than one robot decides to become a leader at the same time, a xed priority among the robots breaks the tie.

Once all the robots have gathered at their starting locations and the leaders have emerged, the preconditions for the overwatch behavior set are satised in all of the robots, causing all the robots to begin watching out for some sort of danger, such as the presence of an enemy agent. (In this simulation, however, no sources of danger were modeled.) The precondition for the lead-to-waypoint behavior set in a leader robot is that the previous team has just bounded to its next waypoint. In order to initiate the bounding at the beginning of the mission, this condition is hardcoded into the leader of the rst team as soon as its team has collected at the initial gathering

Figure 6-15: A typical run of the bounding overwatch mission (read the snapshots from left to right, top to bottom).

6.4. SUMMARY 177 location. Thus, the leader of the rst team initiates the bounding to the next way-point. This, in turn, satises the preconditions of the follow-leader behavior set in the remaining robots on the rst team. The result is that the leader robot leads the team to the next waypoint while the rest of its team follows along. The members of the second team, in the meantime, have activated their overwatchbehavior sets, and are overwatching the rst team's progress. This scenario is shown in the third frame of gure 6-15.

Once the rst team's leader has arrived at its next waypoint, the completion of the lead-to-waypointis broadcast. Upon hearing this, the leader of the second team's preconditions forlead-to-waypoint are satised, causing it to lead its team to its next waypoint while the rst team overwatches. This continues, as shown in gure 6-15, until the teams reach some prespecied destination.

This illustration describes the basic idea behind the fault tolerant execution of the bounding overwatch mission. One can imagine much more complex versions of this mission that involve numerous roles (such as clearing paths, monitoring the rear of the group, searching for good waypoints, and so forth) that must be carried out by various team members. These roles can be easily and dynamically reallocated among team members with the ALLIANCE architecture when needed due to the failure of robot team members or due to increased requirements of the mission (perhaps due to an attack of enemy forces) in the same way as the leader role is dynamically reallocated in this example.

6.4 Summary

In this chapter, three additional proof of concept implementations of the ALLIANCE architecture have been presented: the box pushing demonstration on physical robots, and the janitorial service and bounding overwatch missions on simulated robots. The box pushing demonstration oers a short, easily understood example of key charac-teristics of the ALLIANCE architecture | fault tolerant control amidst the failure of robot team members, and adaptive control due to changing capabilities of the robot team. The janitorial service application oers an example of a mission involving nu-merous independent tasks that must be carried out, and the ability of ALLIANCE to allow robot team members to select their actions to eliminate duplication of ef-fort. It also illustrates an example in which the L-ALLIANCE architecture can be of particular benet in improving the eciency of the team performance. Finally, the bounding overwatch mission oers an illustration of how several tasks with xed ordering constraints can be solved using ALLIANCE. This architecture oers an easy way to achieve dynamic role transferral in missions involving changes in the environ-ment or in robot team. These applications, along with the hazardous waste cleanup

mission described in chapter 3, illustrate the wide variety of applications for which the ALLIANCE architecture is suited.

Chapter 7