Section: Scientific Foundations

Methodology: “Control to the user, Knowledge to the system”

Thinking of future digital modeling technologies as an “expressive virtual pen”, enabling to seamlessly design, refine and convey animated 3D content, is a good source of inspiration. It led us to the following methodology:

  • As when they use a pen, users should not be restricted to the editing of preset shapes or motion, but should get a full control over their design. This control should ideally be as easy and intuitive as when sketching, which leads to the use of gestures – although not necessarily sketching gestures – rather than of standard interfaces with menus, buttons and sliders. Ideally, these control gestures should drive the choice of the underlying geometric model, deformation tool, and animation method in a predictable but transparent way, enabling users to concentrate on their design.

  • Secondly, similarly to when they draw in real, users should only have to suggest the 3D nature of a shape, the presence of repetitive details, or the motion or deformations that are taking place: this will allow for faster input and enable coarse to fine design, with immediate visual feedback at every stage. The modeling system should thus act similarly to a human viewer, who can imagine a 3D shape in motion from very light input such as a raw sketch. Therefore, as much as possible a priori knowledge should be incorporated into the models and used for inferring the missing data, leading to the use of high-level representations enabling procedural generation of content. Note that such models will also help the user towards high-quality content, since they will be able to maintain specific geometric or physical laws. Since this semi-automatic content generation should not spoil user’s creativity and control, editing and refinement of the result should be allowed throughout the process.

  • Lastly, creative design is indeed a matter of trial and error. We believe that creation more easily takes place when users can immediately see and play with a first version of what they have in mind, serving as support for refining their thoughts. Therefore, important features towards effective creation are to provide real-time response at every stage, as well as to help the user exploring the content they have created thanks to intelligent cameras and other cinematography tools.

To advance in these directions, we believe that models for shape, motion and cinematography need to be re-thought from a user centered perspective. We borrowed this concept from the Human Computer Interaction domain, but we are not referring here to user-centred system design (Norman 86). We rather propose to extend the concept, and develop user-centred graphical models: Ideally, a user-centred model should be designed to behave, under editing actions, the way a human user would have predicted. Editing actions may be for instance creation gestures such as sketching to draft a shape or direct a motion, deformation gestures such as stretching a shape in space, or a motion in time, or copy-paste gestures used to transfer of some features from existing models to other ones. User-centred models need to incorporate knowledge in order to seamlessly generate the appropriate content from such actions. Knowledge may be for instance about developability to model paper or cloth; about constant volume to deform virtual clay or animate plausible organic shapes; about physical laws to control passive objects; or about film editing rules to generate semi-autonomous camera with planning abilities.

These user-centred models will be applied to the development of various interactive creative systems, not only for static shapes, but also for motion and stories. Although unusual, we believe that thinking about these different types of content in a similar way will enable us to improve our design principles thanks to cross fertilization between domains, and allow for more thorough experimentation and validation. The expertise we developed in our previous research team EVASION, namely the combination of layered models, adaptive degrees of freedom, and GPU computations for interactive modeling and animation, will be instrumental to ensure real-time performances. Rather than trying to create a general system that would solve everything, we plan to develop specific applications (serving as case studies), either brought by the available expertise in our research group or by external partners. This way, user expectations should be clearly defined and final users will be available for validation. Whatever the application, we expect the use of knowledge-based, user-centred models driven by intuitive control gesture to increase both the efficiency of content creation and the quality of results.