The Role of Virtual Reality in Autonomous Vehicles’ Safety Alexandre M. Nascimento 1,2 , Anna Carolina M. Queiroz 3,4 , Lucio F. Vismari 2 , Jeremy N. Bailenson 3 , Paulo S. Cugnasca 2 , João B. Camargo Junior 2 and Jorge R. de Almeida Jr 2 1 Center for Design Research (CDR), School of Engineering, Stanford University, California, USA 2 School of Engineering (Poli), University of São Paulo (USP), São Paulo, Brazil 3 Virtual Human Interaction Lab, Department of Communication, Stanford University, California, USA 4 Psychology Institute, University of São Paulo (USP), São Paulo, Brazil E-mail: alexandremoreiranascimento@alum.mit.edu, acmq@stanford.edu, bailenso@stanford.edu, {lucio.vismari, cugnasca, joaocamargo, jorgerady}@usp.br Abstract—Virtual Reality (VR) has played an important role in the development of autonomous robots. From Computer Aided Design (CAD) to simulators for testing automation algorithms without risking expensive equipment, VR has been used in a wide range of applications. Most recently, Autonomous Vehicles (AV), a special application of autonomous robots, became a major focus of the scientific and practitioner community for road and vehicle safety improvements. However, recent AV accidents shed a light on the new safety challenges that need be addressed to fulfill those safety expectations. This paper presents a systematic literature mapping on the use of VR for AV safety and assimilates this literature to create a vision of how VR will play an important role in the development of safety in AV. Keywords—Autonomous vehicles, virtual reality, safety I. INTRODUCTION One of the highest expectations about Autonomous Vehicles (AV) is related to the safety improvements of transportation systems. A significant reduction of the accidents rate is expected by the elimination of the need for a human driver and, consequently, possible human errors. Thus, this expectation relies on the belief the AV will overperform humans in the tasks involved in the process of driving a vehicle. However, the recent accidents involving AVs have demonstrated that the scientific community and industry still have a long path forward before reaching acceptable AV Safety levels. For example, accidents involving Uber’s AV [1] and Tesla’s autopilot [2][3] resulted in life losses. Also, Google has officially reported 272 failures and 13 near misses for its self- driving cars [4]. A recent study about the impact of Artificial Intelligence (AI) on AV safety [5] presented an analysis of 59 studies selected from 4870 initially retrieved papers from multiple indexing databases, using level 5 exclusion criteria defined by [6]. Among the findings, significant gaps were found in AV testing, AV verification and validation (V&V), AV Safety Certification, and safety-oriented AV Design. In other words, the current state of the art about AV is not properly addressing the fundamental safety assurance concepts to the AV deployment. This lack of research could jeopardize the complete adoption of this new promising technology by society. Virtual Reality (VR) is a digital environment that allows users to experience and interact with the environment as if it were real [7]. VR has been used for training (humans) since its inception in the 1960s [8]. In the last decades, VR has been used for training purposes in several fields [9], such as Education [10], Health [11] and Business [12]. VR is particularly useful as it can simulate situations that could be dangerous or life threatening in the real world [13]. Thus, it can provide training possibilities that otherwise would be costly, risky or even impossible in the physical world [14]. Considering the VR affordances, there is a great potential for its application for AV safety as a training, research and test bed tool. VR can be used to train and test interactions between humans and AVs in a realistic environment and without the physical risks of the real environments. Also, it makes the evaluation of users’ behaviors in virtual and controlled scenarios possible. Using VR environments as research, test and training tools allows for greater control of the stimuli the user [15] and the AV’s algorithms are exposed to, and improves the reliability of the research, training and testing, as well as the generalizability of the results. In addition, the use of VR for training AVs’ AI algorithms seems to be a promising and unexplored field. For example, [16] used a simulation and 3D engines for training of deep learning approaches for object classification – a task that usually relies on human annotations. The results indicated the viability of the simulation-only training approach for classifying real world imagery. This kind of training can reduce significant costs and time to generate reliable data sets for AV image recognition. In this context, the present paper intends to deal with the potential of using VR as an instrument to help in filling the gaps on AV safety assurance observed in literature [5] and, consequently, using VR to improve AV safety. After an extensive literature search, no study was found mapping and organizing the related literature to provide a complete vision of the field. Therefore, this paper presents a Systematic Literature Mapping aiming to present a clear picture of the state-of-the-art of the literature in VR on AV safety. II. METHODOLOGY The Systematic Literature Mapping was the selected method for the current study. Based on the process proposed by [17], a 4- steps approach [18] was used in this study: (1) Definition of research questions; (2) Identification of search string and source selection; (3) Study selection criteria; and (4) Data mapping. 50 2019 IEEE International Conference on Artificial Intelligence and Virtual Reality (AIVR) 978-1-7281-5604-0/19/$31.00 ©2019 IEEE DOI 10.1109/AIVR.2019.00017