PROJECT DELIVERABLE D5
SITE INFORMATION
Shared-cost RTD
Project acronym: TOURBOT
Project full title: Interactive Museum Tele-presence through Robotic Avatars
Contract Number: IST-1999-12643
Key Action: 3
Action Line: 3-2-3
TOURBOT: Interactive Museum Tele-presence Through Robotic Avatars
Project Deliverable D5: Site Information
Date Produced: September 27, 2000
Authors: Dirk Schulz, Dirk Haehnel, Wolfram Burgard and Panos Trahanias
Contents
1 Introduction 3
2 Map Construction 4
2.1 Building Grid Maps 4
2.1.1 Laser-based Implementation 4
2.1.2 Maps of the Museum Sites 5
2.2 Map Updating 7
2.3 Maps for User Interaction 8
3 Multimedia Information 11
3.1 Intended Web Interface 11
3.2 Information about Exhibits 12
3.2.1 Foundation of the Hellenic World 13
3.2.2 The Deutsches Museum Bonn 16
3.2.3 The Byzantine and Christian Museum of Athens 20
3.2.4 The Data Collection 25
4 Publications 25
1 Introduction
One of the major tasks of the TOURBOT system is to navigate safely and reliably through the museum’s premises, controlled by Web users and providing camera images and multimedia information about exhibits. To fulfill this task, the system requires different kinds of site information, e.g. information about the environment the robot is operating in and about the exhibits the robot is showing to the users. This deliverable documents the site information acquired and the methodology employed to obtain the environment information required for the operation of the robotic system.
The site information can roughly be divided into two parts. The first relates to the robotic system, which needs information about the exhibition area, in order to be able to navigate safely and reliably in it. For mobile robots, this information is generally provided in the form of a map of the environment, which the robot uses to keep track of its position within the environment. These navigation maps can be constructed automatically from sensor information that the robot collects within the museum. Therefore, the first part of site information refers to the map building process and the actual maps that were constructed for the three user sites; we document the map-building methodology employed by the TOURBOT system and the workspace maps that were obtained using the system.
The second part of the site information relates to the users of the system. It is the actual information about exhibits, which the system presents to the user through its interface. In general, this can be any kind of multimedia content that can be presented using standard (Web) browsers. For the case of the TOURBOT system, there are specific issues that need to be considered: TOURBOT is an on-line system, since the robotic platform moves through the museum, while a user is observing. Therefore, the system’s interface must continuously inform the user about the robot’s current position and actions. Consequently, maps need to be created which are user-perceivable and which can be used to indicate the robot’s position in the museum workspace. The information about exhibits should be provided just-in-time, e.g. when the robot arrives at the exhibit and looks at it. We will document the maps for user interaction that have been constructed, the multimedia material for the museum sites we compiled, and how this material can be accessed.
2 Map Construction
In order to effectively and safely navigate within the museum, the robot requires an internal model of its environment. The representation of choice for this model is maps. Using a map of the environment, the robot is able to localize itself, e.g. to continuously keep track of its current position within the environment. Such a map, therefore, provides the global coordinate frame of the TOURBOT system. The position of the robot, as well as the location of the exhibits, are specified (as coordinates) within this map.
2.1 Building Grid Maps
TOURBOT employs a probabilistic method, known as Bayesian map building, to construct the maps of the museum sites. A fast real-time variant of this method has been developed by participants of the consortium and an implementation of it has been undertaken for the TOURBOT project. Bayesian map building allows to automatically derive grid maps from sensor data. Grid maps divide the environment into a rectangular grid of equally sized cells, where each cell contains the probability that the corresponding space in the environment is occupied.
Using a probabilistic technique known as Bayesian inference, these probabilities can be determined from proximity sensor data that the robot collects within the environment. Proximity sensors measure the distance to obstacles in the surroundings of the robot. Based on (unreliable) position information provided by the robot’s drive and based on a model of the sensor used, the map building process then computes the probability that a certain cell of the grid is occupied, given the sensor measurement received. Bayesian map building integrates subsequent sensor measurements, in order to obtain a complete grid map of the environment.
2.1.1 Laser-based Implementation
The main sensor used by TOURBOT is a laser range-finder, which measures the distance to obstacles in a horizontal plane. The actual sensor used operates with an angular resolution of 1 degree and provides 180 measurements per scan and about 5 scans per second, that way, covering half of the robot’s surroundings. The device is very accurate with an average measurement error of only 5 cm. Figure 1 illustrates the concepts. The left part of Figure 1 shows one scan taken within the Deutsches Museum Bonn. This scan has been taken by the robot RHINO (see Figure 3), which is equipped with two laser range-finders. The right part of Figure 1 shows a grid map constructed from that single scan using a resolution of 10 cm per cell. The complete map, shown in Figure 5, was integrated from 3365 scans.
Figure 1 A 360 degree laser range scan (left) and the resulting grid map (right).2.1.2 Maps of the Museum Sites
During a TOURBOT meeting in Athens, on March 15-16, 2000, we recorded the laser data for the exhibition rooms at (a) the Foundation of the Hellenic World and, (b) the Byzantine and Christian Museum of Athens. For this purpose, the robotic platform Sam was used (see Figure 2 and Figure 3) that roved through the exhibition areas. Note that, although this robot's hardware differs from the hardware of the TOURBOT system, it runs the same navigation software and the same laser range-finder is used.
Figure 2 The robot Sam collecting laser data in the Byzantine and Christian Museum of Athens.
Figure 3 The robotic platforms used to perform mapping of the user sites; (left) the robot Sam within the Foundation of the Hellenic World; (right) the robot RHINO within the Deutsches Museum Bonn.
Based on the data recorded, the TOURBOT map builder constructed the 2 maps shown in
Figure 4. The resolution of these maps is 10cm per cell. The data for the Deutsches Museum Bonn has been recorded in another session using the robot RHINO (also depicted in Figure 3). The resulting map is shown in Figure 5.
Figure 4 Navigation grid maps; (left) Byzantine and Christian Museum of Athens, (right) Foundation of the Hellenic World.
Figure 5 Navigation grid map for the Deutsches Museum Bonn.
2.2 Map Updating
The map building module, as described so far, is intended to construct static maps, e.g. they show the museum environment as it appears when the museum is closed to visitors. However, during the operation of the TOURBOT system, the museums will be open to the public and visitors will be walking around. In order to cope with the dynamic changes within the environment, for example to plan detours around groups of people, the robot must be able to maintain its map over time. Therefore, the TOURBOT system implements on-line map updating. It continuously integrates recent laser data into the static map to obtain maps which provide snapshots of the current state of the robot’s surroundings. In these maps, groups of people and other objects (obstacles) appear as occupied space enabling the robot to plan detours.
2.3 Maps for User Interaction
In its normal mode of operation, the TOURBOT system is controlled by inexperienced users, either remote (Web) users or on-site users. To keep the user informed about what the robot is currently doing, the actions of the robot must be reflected in the user’s Web interface. The robot’s localization component continuously maintains its current position within the museum based on the grid map. In principle, these maps can be used inside the Web interface to communicate the robot’s current position to the user. However, non-technical users are not familiar with concepts and terminology from the field of robotics and, moreover, maps built from sensor data are not particularly intuitive. At the same time, they do not contain any information about the location of exhibits and of any objects that can not be detected by the robot’s sensors.
For user interaction, we therefore use CAD maps, which resemble well known floor plans. These plans are aligned with the grid maps, e.g. they have the same size and can also be used to indicate the robot’s position to the user; additionally, they are annotated with information regarding the position of the exhibits. Figure 6, 7 and 8 show the maps for user interaction for the Foundation of the Hellenic World, the Deutsches Museum Bonn and the Byzantine and Christian Museum of Athens, respectively. It is quite straightforward to compare these maps with the navigation maps that have been autonomously built from sensor data. The similar structure of the two kinds of maps is evident and also differences are quite obvious. The navigation maps (machine oriented) are clearly probabilistic in nature and they show several exits from the exhibition area, which the robot will never use. The latter might confuse the Web user and are omitted in the interaction map.
The robot navigates autonomously using the navigation maps, which are constantly updated, and the robot’s current position is permanently displayed inside the user interaction map using Java applets. In this context, the need arises to explain the robot’s course of action to the Web user. For example: “Why does the robot take this detour, although there is free space in the map?”. For this purpose, techniques to estimate the current position of objects and of people walking around have been developed. This information will be used by the TOURBOT system to update the user interaction map accordingly. It is our goal to display not only the robots current position, but also the detected changes within the environment.
Figure 6 Map for user interaction of the Foundation of the Hellenic World indicating the position of the 22 exhibits.
Figure 7 Map for user interaction for the Deutsches Museum Bonn, indicating the 26 exhibits.
Figure 8 Map for user interaction for the Byzantine and Christian Museum of Athens.
3 Multimedia Information
The major task of the TOURBOT system is to present museum exhibits to visitors, either over the Web or on-site. Web users observe the exhibits through the robot’s cameras and, moreover, get additional information pertinent to the exhibits. TOURBOT achieves this by combining the advantages of a mobile platform with the Web’s capability to transmit and display almost any kind of multimedia material. In this section we present the compilation of the system’s information base, consisting of multimedia material from the three end-user (museum) sites.
3.1 Intended Web Interface
The presentation of multimedia material is innately coupled with various aspects of the Web interface. Although the final design of the TOURBOT Web interface has not settled yet (interface design is a subject for work package 7), it is clear from our earlier experiences with web-controlled robots that, for the most part of an application, the interface must be contained within one web-page. As a consequence, robot-control and information feedback share the same page, putting thus restrictions on the possible presentation of multimedia information.
Figure 9 shows a first version of the TOURBOT Web interface. The page is divided in two parts (left part and right part); in addition, self-explanatory buttons are used for analogous tasks. The lower left part contains the robot control interface; the upper left part is the video window where the user observes the viewed scene. The right part is used by the TOURBOT system, to display just-in-time information about exhibits. Just-in-time is interpreted in our case that information about an exhibit will show-up as soon as the robot arrives at it. The user will then be able to browse through the information on the exhibit.
Figure 9 Prototypical TOURBOT Web interface. The lower left part is dedicated to robot control; the upper left part contains the video window. The right part is dedicated to the presentation of multimedia information.
3.2 Information about Exhibits
The above described concept (and initial implementation) of the user interface has influenced the compilation of multimedia information. In the following, we describe the data collected for each of the three museums and we give a few sample examples. The complete site information is available on a CD-ROM. The structure of the data collection is also described in this section.
3.2.1 Foundation of the Hellenic World
The exhibition in FHW’s workplace (see Figure 6) contains 22 showcases (exhibits). For each of the exhibits, an image and a descriptive text are contained. In addition, related context information is included in form of short texts and images relevant to the specific exhibit. For six exhibits, short video clips are also provided. Currently, the database for the Foundation of the Hellenic World contains 32 descriptive texts, 71 JPG-images, 10 video clips and 1 sound file. The music contained in the sound file has been composed especially for the exhibition. It can be played-back for all the exhibits, but especially in combination with the video clips, which mainly show dances inspired by the exhibits. Figures 10 and 11 constitute a sample example of the visual and textual information associated with exhibit 2.