Goals of the Experimentation

The first goal of the real experimentation is to check whether the good results obtained through simulation are also obtained with the real robot. Ideally this would be the case, so the only modifications needed would be to make the existing Navigation system use the real robot instead of a simulated one. However, moving from simulation to the real world is not that easy, as many problems arise when working with physical robots which were not present on the simulated world (unless the simulator used has very high realism). These problems are mainly related to the motion and vision systems of the real robot.

Regarding the motion system, we have to take into account that the robot needs some time in order to execute motion commands. On simulation, we could run the system as fast as we liked, since the commands were executed immediately, however, we cannot do so with the real robot. The frequency of sending these motion commands to the robot should be set according to the response time of the robot, so a command is only sent when the robot is really prepared to execute it.

Another problem of using a real robot is the vision system. Although the vision system and the landmarks we have designed are very simple, the system is not able to identify the landmarks all the time, due to changes in illumination, interference on video transmission, blurring caused by motion of the camera, etc. Therefore, as already mentioned, the vision system needs to process some frames before it is able to inform about the detected landmarks. Thus, the actions for moving the camera and identifying landmarks must also be sent with the proper frequency so that the vision system has time to process enough frames.

To overcome these problems, we have tuned the agents so that the robot is able to execute all the commands generated by the system.

Through the real experiments we also check whether the Navigation system we have designed is able to perform well in different types of environments, and if the design of each individual agent is the most appropriate for obtaining good overall performance of the Navigation system. To check this, we have experimented with different scenarios, starting with simpler ones and increasing their complexity step by step. The two main variables that describe the complexity of a scenario are:

• Density of landmarks: the fewer landmarks in the scenario, the more risky it is, since the map contains very little information about the relative location of the target and other landmarks. On the other hand, if the density of landmarks is high, there will very probably be always some landmarks visible, and the Navigation system will be able to compute the location of the target from the visible landmarks.

• Density of obstacles: if the density of obstacles is low, the path from the starting point to the target may not be blocked, or only blocked by easily avoidable obstacles, so the robot may not need to compute diverting targets to reach the original one. Contrarily, in a scenario with many obstacles, the robot is forced to change direction very often, which may cause it to lose sight of the target, and therefore, to increase the imprecision about its location. Moreover, if the obstacles block the way to the target, the Navigation system may need to compute a diverting target to reach the original one.