ASSESSMENT AND DIAGNOSIS FOR VIRTUAL REALITY TRAINING Stéphane Ploix, Matthieu Désinde, Florence Michau (Stéphane.Ploix@inpg.fr, Matthieu.Desinde@inpg.fr, Florence.Michau@inpg.fr) Laboratoire d’Automatique de Grenoble, INPG, UJF, UMR 5528 BP 46, F-38402 Saint Martin d’Hères Cedex, France Phone: 33 4 76 82 62 44, Fax: 33 4 76 82 63 88 KEYWORDS: Logical diagnosis, virtual reality training environment, learner's behaviour modelling, system theory Abstract: Simulators and virtual environments are powerful tools to train people in “reality- like” situations. Regardless of application context, learner actions are currently monitored by alert functions embedded into software environments that immediately detect misbehaviours and misconceptions. Nevertheless, detection does not make it possible to point out the possible inabilities of learners. Embedding diagnostic procedure is a great issue to improve lots of training environments. This paper presents a modelling of learner’s operative tasks suitable for virtual reality learning software. This modelling is used to design behavioural assessments and diagnostic analyses. INTRODUCTION Real-life experience is the best way to learn. Unfortunately, this is often too expensive, too dangerous or simply not possible to carry out. Simulators and virtual environments are powerful tools to train people in “reality-like” situations. They represent an increasing economic challenge in a large number of application fields such as operator training, surgeon training or more commonly driver or pilot training. Regardless of application context, learner actions are currently monitored by trivial alert functions embedded into software environments that immediately detect some well identified misbehaviours and misconceptions (software parameters exceeding a validity range, failure to comply with instructions, dangerous situations, etc...). Detection in complex situation, such as virtual reality, is difficult to handle with such empirical tools. Moreover, detection does not make it possible to point out the possible inabilities of learners. Embedding diagnostic procedure is a great issue to improve lots of training environments. This paper presents a modelling of learner’s operative tasks suitable for virtual reality learning software. This modelling is used to design behavioural assessments and diagnostic analyses. After a short description of the addressed issues, the proposed modelling is detailed and methods for assessment and diagnostic analysis are depicted. The modelling and methods proposed have been applied to a real application dealing with “child safety while walking in the street” using a virtual reality training environment named RESPECT©. RESPECT software is developed by CORYS TESS Company and different research labs: INRETS/LPC, UVHC/LAMIH and CRP2C, with the support of the French Ministry of Youth, Education and Research to train children for respecting safety pedestrian rules in a 3D video game like environment. Children move in a 3D street representation. They have to reach a target according to safety and traffic rules. The presented results aim at improving the assessment and the analysis of learner abilities, which were formerly limited to few alert functions. Whatever the application fields are (monitoring and supervision of complex systems (nuclear power plant, cement works, refinery, production line, recovery room), mobile control (trains, aircraft), traffic control or crisis management (setting up emergency plans)), assessment and diagnosis purposes in the framework of virtual reality training leads to the same main issues: the