SIMULATION AND VISUALIZATION OF INDOOR-ACOUSTICS FOR ROBOT CONTROL Norbert Schmitz, Jens Wettach, and Eduard Deines IRTG, Dep. of Computer Science University of Kaiserslautern Kaiserslautern, Germany email: {nschmitz|wettach|deines} @informatik.uni-kl.de Peter Dannenmann and Martin Bertram German Research Center for Artificial Intelligence P.O. Box 2080 Kaiserslautern, Germany email: {Peter.Dannenmann| Martin.Hering-Bertram}@dfki.de Karsten Berns and Hans Hagen Department of Computer Science University of Kaiserslautern P.O. Box 3049 Kaiserslautern, Germany email: {berns|hagen} @informatik.uni-kl.de ABSTRACT An autonomous robot orientates itself by using informa- tion provided by its sensor systems. Besides distance sen- sors, optical and acoustic sensors play a vital role in fulfill- ing this task. In this paper we present a novel approach to simulate and visualize the acoustic properties of an indoor scene. This simulation data is used for testing and refining the control algorithms of an autonomous robot interacting with humans in an office environment. In order to enable a robot to interact with its environment and to perform its intended tasks in a context-sensitive manner, it must be ca- pable of interpreting the information provided by its sensor systems. For testing these interpretation capabilities, cer- tain environmental stimuli need to be provided to the robot in a controlled and repeatable manner. However, such stim- uli are not available normally. The presented work provides a virtual testing environment that permits the realistic simu- lation and visualization of the acoustic properties of indoor- environments. Thus, it is possible to simulate and visualize the spread of sound waves within a room and to simulate the acoustic signals a robot receives at certain positions. Using such well-defined test conditions and a visual repre- sentation of the spread of sound in the test environment, it is now possible to find positions of special acoustic properties and to use them for testing the robot’s reactions to acoustic events. On this basis, the robot’s control algorithms can be refined accordingly. KEY WORDS Virtual Environments, Acoustics, Simulation, Visualiza- tion, Robotics, Control Algorithms 1 Introduction Mobile robots can increasingly often be found in private houses. They either mow the lawn or vacuum the floor just to mention a few examples. These robots still lack some important abilities as the ability to communicate. The first step to man-machine communication is the recognition of humans by their voices. There are other sources of sound besides human beings which should also be recognized. A patrolling robot would also be interested in the sound of breaking glass, e.g.. Although different sound localization systems do ex- ist, it is difficult to compare or optimize those algorithms. A simulation of sound source and sink can generate repeat- able test scenarios. The overall performance of such a lo- calization system depends on reliable tests with a realistic simulation engine. In this paper we will introduce such a simulation system for mobile indoor robots. In the next section we give a short overview of the cur- rent state of the art in robot navigation as well as in acous- tic simulation and put this into relation to previous work performed at our research groups. In section 3 and 4 we then give a more detailed description of the technologies of sound localization and simulation of acoustics, that we have developed. Finally, we describe the virtual acoustic test environment for robot control, that is currently under development. 2 Related Work 2.1 Current Research Work in Robot Navi- gation and Acoustic Simulation Modeling the application environment of a mobile robot can be based on maps of different types. On the top level of abstraction a topological map represents interesting po- sitions that can be recognized by the robot’s sensor system as well as their spatial relationship [1], [2]. In grid maps the environment is divided into two-dimensional cells which are marked with attributes like occupied, free or unknown. Thus they are primarily used for navigation and obstacle avoidance. Geometric maps contain exact dimensions of objects represented by lines, polygons or rectangles. Fur- thermore, distance information between objects is recorded [2], [3]. Additionally, there exist several combinations of these three types of maps, for example to fuse the accuracy and consistency of grid maps with the efficiency of a topo- logical map [4]. These maps describe the structure of the environment which has great influence on the simulation of sound. Simulation of acoustic properties of rooms can be per-