A Trial Example

In this section we describe in detail one of the trials run in a scenario of class 4. The environment and the path followed by the robot are shown in Figure 6.9. The target landmark in this trial is landmark number 10. In Figures 6.10 and 6.11 the incremental building of the map is depicted. They show both a 2D representation and the topological map actually stored by the Map Manager. In the topological maps, although not shown, the arcs have a fixed cost of 1, unless otherwise specified. Figures 6.12 and 6.13 show the evolution of the bids of each agent and the Pilot for moving and looking actions, respectively. In these graphics, the filled areas indicate the agent that made the highest bid at that point in time. The corresponding points in Figure 6.9 are also shown. Next, we comment on the relevant points of the path:

Figure 6.9: Path followed during the trial. See explanation of relevant points on the text

Figure 6.10: Map created during the trial

Figure 6.11: Map created during the trial (cont.)

Figure 6.12: Moving bids. Target Tracker in red, and Pilot in green

Figure 6.13: Looking bids. Target Tracker in red, Pilot in green and Risk Manager in blue

• A: Starting point of the trial. Initially, landmarks 10, 29 and 19 are visible. With these three landmarks, no map is created, since at least four landmarks are needed in order to start building the map. Landmark 10 is selected as the target by the user and the Rescuer is informed about it. Then, the Rescuer bids for doing an initial sweep, as described in Section 4.4.4. During this sweep, landmarks 4, 21 and 17 are also identified. With these new landmarks, the Map Manager is able to start building the map. The step by step update of the map is shown in Figure 6.10. The corresponding updates after seeing each of these three landmarks are maps (1) to (3). When the sweep is finished, the Rescuer informs the Target Tracker about the target being landmark 10, which immediately starts bidding for going towards it, and the robot starts moving. Actually, the point A in the graphics of the bids corresponds to this moment, when the Target Tracker starts bidding. As can be seen in the graphic of moving action bids, the Pilot won most of the bids. This was so because landmark 4 was close to the robot, and the Pilot wanted to avoid it. The trajectory, however, was minimally modified. Before reaching point B, landmark 13 is identified, and the map is updated accordingly, resulting in map (4) in Figure 6.10.

• B: The robot bumps into the obstacle between landmarks 29 and 4 and immediately backs up. However, it is not yet considered as being a long blocking obstacle, since there is still enough space between the crash point and landmark 29, through which the robot could pass. This back up is a built-in action of the Pilot, and it does not bid for executing it. That is why in the graphic the Target Tracker wins the bids. However, while the back up action is being executed, these bids are not taken into account.

• C: After backing up, the Target Tracker bids again for moving towards the target, but these bids are surpassed by the Pilot's bids to avoid the just detected obstacle (as can be seen in the moving bids graphic), and the trajectory is slightly modified.

• D: The robot bumps again into the obstacle and backs up. After this second crash, the obstacle is considered to be blocking the path. The Pilot informs the Navigation system about the blocking situation. This information is internally sent to the Map Manager, which updates the map (the corresponding arc is assigned an infinite cost, see map (4b) in Figure 6.11), and to the Rescuer, which asks the Map Manager for a diverting target. Again, although in the graphics the Target Tracker is winning the bidding, the back up action is really being executed.

• E: The Map Manager computes the diverting target as being: ``to cross the edge between landmarks 17 and 29'' and informs the Rescuer, which will inform the Target Tracker about the new target. This agent starts bidding to move the robot so that it crosses the given edge.

• F: At this point, the Target Tracker considers that the edge 17/29 has been crossed and informs about it. This causes the Rescuer to set the target to be the original one (landmark 10). The Target Tracker's bids are again to move towards this landmark. Before reaching point G, landmarks 1 and 20 are detected and the map is updated (maps (5) and (6)). Landmark 20 is not visible in Figure 6.9; it is behind landmark 1.

• G: The proximity of landmark 13 makes the Pilot bid high to avoid it, surpassing the Target Tracker's bids, and the robot's trajectory is modified. While avoiding this landmark, landmark 7 is detected, and the map is updated, resulting in the final map (7).

• H: At this point the Pilot considers that landmark 13 has been avoided and stops bidding. The Target Tracker wins again, and it makes the robot go towards the target.

• I: The target is finally reached.

Analyzing the graphic of looking action bids, we can see that the winning bid is periodically changing between the Pilot and the Risk Manager. The bids of the Target Tracker are very low, since the target is precisely located during the whole trial. Around point H, the bids of the Risk Manager also decay. This is so because at that point, there are more than six landmarks behind the robot, which makes the risk 0. The winning bids of the Target Tracker, at point I, are due to the fact that this agent bids very high to look towards the target when this has been reached. The execution of this action has no intention of decreasing the imprecision of the target's location, but it is just a way to show that it ``knows'' that the target has been reached.