Normatividad Técnica, Sanitaria y Comercial

the Integration of the Object Attention System

The use of real robots for the development of the Object Attention System is essential mostly for two reasons. First, it enables a more robust implementation as it allows to evaluate the proper functionality of the Object Attention System. Secondly, the use of real robots demonstrates that the system not only operates in a laboratory environment, but actually on the intended field of application. While in the beginning of the development phase of the Object Attention System only one mobile platform existed, later on a stationary anthropomorphic robot became available as well. As a consequence, the proposed Object Attention System has been enhanced for the use with this robotic platform as well as it partly offers different sensors. Thus, the Object Attention System can be used with human- like sensors only and does not necessarily rely on artificial sensors, like the laser range finder any longer. The following paragraphs, therefore, briefly describe the different hardware used.

The Mobile Robot BIRON

One of the first steps during the development of the Object Attention System con- sisted in the examination of the existing robot platform with regard to the needs for the integration of the Object Attention System in the robot. It turned out

4.2. Hardware Platforms used for

the Integration of the Object Attention System 57

that for several reasons, e.g., insufficient computational power, it did not match the requirements for a successful integration. Thus, an enhanced new platform has been assembled. As a consequence, not only the Object Attention System but also further modules, like the gesture recognition by N. Hofemann [HHFS05] could be integrated as well. Image 4.7 shows the new mobile Bielefeld Robot Companion (BIRON). It is based on a Pioneer PeopleBot platform from the com- pany ActivMedia. Next, this hardware platform is described in more detail.

Figure 4.7: The Bielefeld Robot Companion (BIRON) of the applied computer sci- ence group at Bielefeld University. The wirelessly connected notebook depicted on the right is mainly used for speech processing tasks, e.g., the dialog system.

On its top, a Sony Evi-D31 pan-tilt camera is mounted at the front side at a height of approximately 1,42 m. It is the most important visual sensor for the Object Attention System as it is used to identify objects and, therefore, it is described more detailed than the other sensors. Besides, it is also used to identify the face of the current interaction partner (cf. 3.1 on page 34). The camera supports a resolution up to 768×576 pixel (PAL) and is steerable for about 100◦ to the left and as well as to the right, and for about 25◦ in each vertical direction. Its lens provides a field of view of up to 37,6◦× 48,8◦ _{in its widest zoom position while the} maximal ×12 optical zoom provides a perceptional field of view of 3,2◦_{× 4,3}◦_{. The} zoom functionality is approximately linearly adjustable over an interval divided in approximately 1000 steps. The camera is attached to a mounting that includes a display with a touch screen. This display is used for maintenance tasks as well as for visual feedback for the user during a Human-Robot Interaction. At the left and right bottom of the display two far-field microphones are mounted which capture the environmental sound for the speaker localization as described by S. Hohenner in [Hoh05]. Additionally, in a calm environment these microphones can be used

58 4. Development of an Object Attention System to capture the user’s speech which makes the use of a close-talking microphone headset dispensable.

Beneath the display, a sonar ring including 8 sensors is mounted while an equally equipped second sonar ring is mounted at the base of the robot. These sensors can be used for, e.g., collision avoidance, but as they produce a clicking sound in audible frequency they are disabled as they would influence the speech recogni- tion results. Directly under the upper sonar ring, a monocular Apple iSight camera is mounted which is used for the body tracker and the gesture recognition, cf. sec- tion 3.1 on page 36 and the following ones. Alternatively, a stereo camera can be mounted in order to get more accurate depth values. Also embedded into the tower casing of BIRON are two speakers that are used for its utterances during an interaction phase. Besides, the tower also contains an industrial computer that is mainly used for the visualization on the display and the Person Tracking and Attention System by M. Kleinehagenbrock [Kle05], S. Lang [Lan05], and T. Spexard [Spe05,SHFS06]. At the bottom of the tower casing two mountings are attached, one on each side in order to equip the robot with two notebooks that perform the computations of the Object Attention System, the gesture recogni- tion, and a self-localization which is currently under development. Between the two laptops, the blue-colored laser range finder is shown. This 2D-laserscanner measures distances of the environment within a 180◦ plane scan at an approxi- mate height of 30 cm. Thus, pairs of legs can be detected by the Person Tracking and Attention System by using an appropriate heuristic as described by S. Lang and J. Fritsch in [Lan05,FKL+_03].

The basis of BIRON holds a motor which enables BIRON to move forward and to rotate. Since no sensors are oriented to the back, a backwards movement is disabled and so are the black bumper switches at the very bottom that can, for instance, be used for emergency stops if BIRON collides with an obstacle. Within the basis, the power supply consisting of high-capacity batteries and a second industrial computer are contained. This computer is connected to the two microphones beneath the display and to the motor controller board. Additionally, it serves as host computer for the wired network on the robot and the wireless network that is used to connect one or more notebooks that are used for, e.g., the speech processing as shown in Figure4.7.

The technically oriented platform BIRON is due to its sensors not ideally designed to simulate the behavior of humans and, therefore, restricts the fields of applica- tion of the Object Attention System. Thus, the Object Attention System has been extended for the use with anthropomorphic robot platforms as well.

The Anthropomorphic Robot BARTHOC

In this section, the anthropomorphic robot Bielefeld Anthropomorphic Robot for Human-oriented Communication (BARTHOC) [HSF+_{05,SHFS06] is described. It} has been developed by the company mabotic and is mainly used for user studies with human-like robots. The applied computer science group of Bielefeld Univer- sity has got two exemplars of BARTHOC, an adult-like one (BARTHOC senior ) shown in Figure4.8(a) and a second one with child-like dimensions (BARTHOC junior ), see Figure 4.8(b). The latter one is shown with a skin-like mask for the face while an appropriate mask for BARTHOC senior exists as well.