AUGMENTED REALITY FOR INDUSTRIAL ASSEMBLY GUIDANCE: HOW EFFECTIVE IS IT

Augmented reality (AR) is expected to play an increasing role in numerous professional fields, such as medicine, aviation, and especially in industry. Given its potential to improve training in assembly task, many studies have tried to assess the effectiveness of AR using various objective or subjective criteria. The aim of this article is to review what is currently known about the use of AR for assembly training. The first part of this paper presents a definition of AR and the different AR devices that are currently proposed. The second part briefly describes the different applications of AR and its interest for industrial assembly processes. In the third part, we propose a literature review of studies published over the last 20 years, on the effectiveness of presenting assembly instructions using AR. Our analysis suggests that effectiveness is mainly evaluated using three criteria: time to complete the assembly, number of assembly errors and mental workload of the user. Data from these studies show that the presentation of assembly instructions in AR can reduce the number of assembly errors, but is not systematically effective in shortening the time of the overall assembly task or in reducing the mental workload of the user. Thus, there is currently no consensual conclusions about the effectiveness of AR for training. In the last part of this article, we suggest that this lack of conclusive data could be explained by a lack of unified and comparable methodology between studies. We identify and discuss four factors that differ between the reviewed studies, such as heterogeneity in the population studied, the type of assembly task used, the AR systems tested but also evaluate AR for guidance or learning purpose. In addition, most of the studies evaluated test situations that are not representative enough of an industrial context. In conclusion, we emphasise that further studies on the impact of AR integration for training directly in industry are needed. Without more representative data on AR, its integration must be done with caution and could not completely replace a human instructor.