EN FR
EN FR


Section: New Results

Fundamentals of Interaction

Participants : Michel Beaudouin-Lafon [correspondant] , Wendy Mackay, Cédric Fleury, Theophanis Tsandilas, Dimitrios Christaras Papageorgiou, Han Han, Germán Leiva, Nolwenn Maudet, Yujiro Okuya, Miguel Renom, Philip Tchernavskij, Andrew Webb.

In order to better understand fundamental aspects of interaction, ExSitu conducts in-depth observational studies and controlled experiments which contribute to theories and frameworks that unify our findings and help us generate new, advanced interaction techniques. Our theoretical work also leads us to deepen or re-analyze existing theories and methodologies in order to gain new insights.

Continuing our long-standing exploration of Fitts' law, we demonstrated the dangers of confounding factors in Fitts'-like experimental designs and recommended how to avoid them [20]. Confounds come from the fact that traditional Fitts'-like experiments use geometric progressions of the two main factors (target distance D and amplitude W) and aggregate data points per ID=log(1+D/W). This typically leads to a strong confound between D and ID, whereby an effect attributed to ID may in fact be due solely to D. We showed evidence of published results where this confound led to the misinterpretation of experimental results, and proposed stochastic sampling of D and W as a technique to avoid such problems.

We also reviewed statistical methods for the analysis of user-elicited gestural vocabularies [16] and argued that current statistics for assessing agreement across participants are problematic. First, we showed that raw agreement rates disregard agreement that occurs by chance and do not reliably capture how participants distinguish among referents. Second, we explained why current recommendations on how to interpret agreement scores rely on incorrect assumptions. Third, we demonstrated that significance tests for comparing agreement rates, either within or between participants, yield large Type I error rates (>40% for α=.05). As alternatives, we presented agreement indices that are routinely used in inter-rater reliability studies. We discussed how to apply them to gesture elicitation studies. We also demonstrated how to use common resampling techniques to support statistical inference with interval estimates. We applied these methods to reanalyze and reinterpret the findings of four gesture elicitation studies. We also participated in an invited formal debate at ACM/CHI 2018 to discuss the issue of replicability in HCI experiments, specifically whether or not the community should adopt the TOP (Transparency and Openness) guidelines for data and code transparency, citation, experiment preregistration and replication of experiments.

In order to explore novel forms of interaction based on the concepts of interaction instruments and interactive substrates, we conducted several studies and developed prototypes in three main areas:

First, we challenged the notion of application as the main organizing principle of digital environments. Most of our current interactions with the digital world are mediated by applications that impose artificial limits on collaboration among users and distribution across devices, and the constantly changing procedures that disrupt everyday use. These limitations are due partly to the engineering principles of encapsulation and program-data separation, which highlight the needs for appropriate conceptual models of interaction [18]. We proposed new architectural principles [28], [17] that address these issues by considering interactions as first-class objects that can be dynamically created, added to and removed from an interactive system.

Figure 2. Montage: the UserCam captures the context (a) and the WizardCam captures the paper prototype (b); Both live-stream video to the Canvas, where the designer can add digital sketches (c). Montage replaces the green screen with the interface to create the final composition (d).
IMG/Montage.png

Second, we addressed the needs of designers and developers of interactive systems through a series of studies and prototypes. Current prototyping tools do not adequately support the early stages of design, nor the necessary communication between designers and developers. We created and evaluated VideoClipper and Montage [21], two tools that facilitate video prototyping for the early sketching of ideas. VideoClipper facilitates the planning and capturing of video brainstorming ideas and video prototypes, while Montage (fig. 2) uses chroma-keying to create more advanced video prototypes and facilitating their reuse in different contexts. We also created Enact (under submission), a prototyping tool that lets designers and developers work in the same environment to create novel touch-based interaction techniques. Germán Leiva, supervised by Michel Beaudouin-Lafon, successfully defended his Ph.D. thesis Interactive Prototyping of Interactions: From Throwaway Prototypes to Takeaway Prototyping  [34] on this topic.

Figure 3. Pre-computed meshes of a rear-view mirror while modifying the right part: the user's hand position (Phand) determines the selected shape (left). A virtual car cockpit where the user modifies the rear-view mirror shape in real time, using haptic force feedback (right).
IMG/shapeguide.png

Third, in the context of Computer Aided Design (CAD), we explored solutions for modifying parametric CAD objects in an immersive virtual reality system. In particular, we developed ShapeGuide [14], a technique that lets users modify parameter values by directly pushing or pulling the surface of a CAD object (Figure 3). Including force feedback increases the precision of the users’ hand motions in the 3D space. In a controlled experiment, we compared ShapeGuide to a standard one-dimensional scroll technique to measure its added value for parametric CAD data modification on a simple industrial object. We also evaluated the effect of force feedback assistance on both techniques. We demonstrated that ShapeGuide is significantly faster and more efficient than the scroll technique. In addition, we showed that force feedback assistance enhances the precision of both techniques.