Naughty Agents Can Be Helpful: Training Drivers to Handle Dangerous Situations in Virtual Reality Yongwu Miao*, Ulrich Hoppe*, and Niels Pinkwart** * Institute for Computer Science and Interactive Systems The University of Duisburg-Essen, Germany (miao, hoppe)@collide.info ** Human Computer Interaction Institute Carnegie Mellon University, USA nielsp@cs.cmu.edu Abstract Recently many car driving simulators have been developed and used for educational purposes. They can provide special training opportunities that are not available frequently in reality. However, currently existing simulators often rely on predefined situations and scenarios. Training students to deal with spontaneously occurring dangerous traffic situations is difficult when using hard wired “emergency” situations that occur in a predictable manner. In this paper we present a new approach to provide special learning opportunities. We adopt intelligent agent technologies to develop the “problem creator”, a pedagogical agent who can deliberately cause dangerous situations. Introducing the problem creator into our collaborative 3D virtual car driving environment makes it unpredictable when, where, and what type of abnormal situations the students will be confronted with. These irregularly occurring challenges model the reality of car driving well. 1. Introduction Car driving is a kind of high-performance task where the driver must continuously adjust his behavior to changes in his environment. Sometimes, a driver has to make decisions and take actions quickly to handle unfamiliar and even dangerous situations that occur spontaneously. Without any skill to react to such an urgent event once it occurs, the driver may cause (or at least be involved) in serious accidents and even pay his life. Therefore, training to handle (and intuitively anticipate) abnormal situations should be a very important part in car driving training programs. However, it is currently impossible or too expensive to set up such training situations in reality. The advantages of simulators for training high- performance tasks like car driving have been pointed out in literature [5, 10]. Using a simulator, a trainee is able to practice the handling of potentially dangerous emergency situations in virtual reality. Because these “emergency scenarios” are usually hard coded or scripted [3, 11], a simulator can repeatedly present traffic situations to students, replay driving failures to learners, and repeat exercises in a training course until the trainee can handle them successfully. Yet, the problem is that if the same “emergency situation” occurs repeatedly, a trainee can easily predict what will happen and thus prepare for the situation. This is unrealistic, since in the reality of car driving emergency situations occur unpredictably. In this paper, we present a new approach to train learners to handle abnormal situations. We employ the problem creator, a kind of pedagogical agent that can deliberately create abnormal situations within a collaborative 3D car driving simulation environment. When, where and in which order the abnormal situations will be created is unpredictable. In this paper, we will give an overview of our collaborative 3D car driving simulation. Then the design and implementation of a problem-creator will be described. We also discuss the usage of the problem-creator and compare related systems. Finally, we draw conclusions of our work and point out future directions. 2. A Collaborative Car Driving Simulation Environment This section provides an overview of our collaborative 3D car driving simulation environment. In this learning environment, multiple users can drive cars in a shared virtual driving place. Thus, specific traffic situations can involve the cars of multiple