Towards a Model for Predicting Intention in 3D Moving-Target Selection Tasks

Abstract

Novel interaction techniques have been developed to address the difficulties of selecting moving targets. However, similar to their static-target counterparts, these techniques may suffer from clutter and overlap, which can be addressed by predicting intended targets. Unfortunately, current predictive techniques are tailored towards static-target selection. Thus, a novel approach for predicting user intention in movingtarget selection tasks using decision-trees constructed with the initial physical states of both the user and the targets is proposed. This approach is verified in a virtual reality application in which users must choose, and select between different moving targets. With two targets, this model is able to predict user choice with approximately 71% accuracy, which is significantly better than both chance and a frequentist approach. KeywordsDepartment of Psychology, Virtual Reality Application Center, User intention, prediction, Fitts' Law, movingtarget selection, perceived difficulty, decision trees, virtual reality Abstract. Novel interaction techniques have been developed to address the difficulties of selecting moving targets. However, similar to their static-target counterparts, these techniques may suffer from clutter and overlap, which can be addressed by predicting intended targets. Unfortunately, current predictive techniques are tailored towards static-target selection. Thus, a novel approach for predicting user intention in movingtarget selection tasks using decision-trees constructed with the initial physical states of both the user and the targets is proposed. This approach is verified in a virtual reality application in which users must choose, and select between different moving targets. With two targets, this model is able to predict user choice with approximately 71% accuracy, which is significantly better than both chance and a frequentist approach.

Using relative head and hand-target features to predict intention in 3D moving-target selection

“…Taking this idea to the extreme, Casallas and colleagues [5] Expanding on the work from [5], the present work introduces relative head-target and hand-target features calculated during a time window, to predict intended target. These features can be generalizable to different moving-target selection tasks.…”

“…Additionally, these features are integrated with the previous initial-target-state model [5] to demonstrate enhanced predictive performance.…”

“…A possible solution to these limitations, also present in static selection, is to predict the intended targets [12,20]. Unfortunately, to the authors' knowledge, most predictive techniques are tailored toward static-target selection, except for [5,29].…”

“…Based on the promising results of using target size [5] and distance [29] to predict intended moving-targets, the present work explores the integration of head pose with target size and distance to generate a new moving-target predictive model. Due to the inherent visuomotor nature of moving-target selection, it is expected that this model will outperform the existing predictive models.…”

“…Due to the inherent visuomotor nature of moving-target selection, it is expected that this model will outperform the existing predictive models. The scope of this work, however, is limited to the simple, two-target-with-lineartrajectory selection task from [5]. The data from [5] is reevaluated and its predictive model, together with the one from [29] are used as a baseline for the new, proposed model.…”