Bases teóricas de la variable escritura Escritura

Somatic Marker Network

Robot

Modules (C++)

VentralBG

Actions

XML-Configurations

<stimuli>

...

</stimuli>

<actions>

...

</actions>

<rewards>

...

</rewards>

Uses modules

Starts/stops actions

Contain names of actions

Reads information from Use modules

Conf-Tool

Figure 5.1: Overview that shows how all types of users are able to enhance the robot’s capabilities.

the participants having difficulties with the navigation within the tool. These problems can be traced back to a missing overview on the single steps that have to be processed.

A more severe problem was the participants’ confusion about some terms like ’stimulus’. Due to the fact, that there were no or insufficient explanations of the whole learning concept and the used terms, the participants were not able to understand the semantic meaning of the single processing steps.

Based on the preliminary study’s results, a redesigned tool was developed to overcome the previously discussed problems. In the following the different components of the redesigned tool are described. Subsequent to that, an evaluation concerning the usability of the tool is presented.

5.2 Implementation Details

An overview on the tool is given in figure 5.2. The tool is written in Java and uses JavaFX for the graphical user interface. The main menu consists of seven items. In the following, the functionality of each menu item is briefly

Figure 5.2: Start screen of the Conf-Tool. The main area offers information and elements for assistance, while navigation is shown on the left hand side.

described before the evaluation is presented. The tool offers an Expert Mode which can be used by experienced user to configure an application. As this mode did not play any role for the study it is not explained any further.

1. Start: When the program is started, the user gets information about the functional range of the tool and general information about the framework.

This information includes a short description of stimuli, actions and the possibilities of giving the robot rewards in order to influence its behaviour.

Furthermore, the user can start two short videos one of which also gives an overview on the tool, while the other provides a more detailed explanation on how the robot learns.

2. Connection: If one or multiple robots are powered on, the user is able to connect to one specific via the menu item Connection. The tool automatically checks which robots are available and shows them to the user. For more

5.2. Implementation Details 95

. . .

<condition>

<currentValue>LeftBumperPressed</currentValue>

<operator>==</operator>

<value>1</value>

</condition>

. . .

Listing 5.1: Exemplary condition for which a self-documenting alias is existing.

information, the user is able to start a tutorial video which shows how to connect to a robot.

3. Profiles: A profile represents a completely configured application which includes the definitions of recognizable stimuli, executable actions and possi-bilities of obtaining rewards. Therefore, each profile is saved into a separate XML configuration. The user is able to save completed configurations and to load them in order to do some enhancements or modifications. For getting more information about profiles the users can open a tutorial video.

4. Actions: The tool automatically checks which actions are installed on the robot and displays them to the user. The user then is able to choose the relevant actions for the application. Thus, the user defines the set of actions A (see section 2.4). An exemplary entry within the XML file is shown in section 4.4 (see listing 4.4). As for the other menu items, which are described before, a tutorial video is available which can be used by the user to get more information.

5. Stimuli: In addition to the set of actions A, the user has to define the set of stimuli S (see section 4.3). In order to assist the user, self-documenting aliases for different sensors as well as for complete conditions are offered to the user.

For example, the user can choose left hand pressed as a condition which is transformed internally into the XML code shown in listing 5.1. Therefore, the user does not need to know, the name of the sensor’s variable (LeftBumper-Pressed), the relational operator (==) and the sensor’s range of values (0 or 1). Even for this menu item a tutorial video is available to assist the user.

Figure 5.3: Shows the default configuration of the reward possibilities.

6. Rewards: Before the configuration of an application is finished, the user has to define how the robot can be rewarded. If the expert mode is deactivated, a default configuration is used which allows users to give rewards via the tactile sensors on the robot’s head (see figure 5.3). For the study, only the default mode was used, therefore this configuration step was of a purely informative character and did not require any actions by the user.

7. Transmission: In order to start the application, the user is able to transfer the created configuration to the robot. After the transmission is finished, the robot restarts its NAOqi (see section 4.1) and loads the information from the XML file. Furthermore, the user is advised to get the robot into a safe position for the restart. If a specific action called safePosition is available on the robot, this action is executed automatically before the restart.