In the context of optimization and cycles reduction for product design in industry, digital collaborative tools have a major impact, allowing an early stage integration of multidisciplinary challenges and oftentimes the search of global optimum rather than domain specific improvements. This paper presents a methodology for improving participants’ implication and performance during collaborative design sessions through virtual reality (VR) tools, thanks to intention detection through body language interpretation. A prototype of the methodology is being implemented based on an existing VR aided design tool called DragonFly developed by Airbus. In what follows we will first discuss the choice of the different biological inputs for our purpose, and how to merge these multi- modal inputs a meaningful way. Thus, we obtain a rich representation of the body language expression, suitable to recognize the actions wanted by the user and their related parameters. We will then show that this solution has been designed for fast training thanks to a majority of unsupervised training and existing pre-trained models, and for fast evolution thanks to the modularity of the architecture.