At the core of the contribution of this dissertation there is an augmented reality (AR) environment, StaticAR, that supports the process of learning the fundamentals of statics in vocational classrooms, particularly in carpentry ones. Vocational apprentices are expected to develop an intuition of these topics rather than a formal comprehension. We have explored the potentials of the AR technology for this pedagogical challenge. Furthermore, we have investigated the role of physical objects in mixed-reality systems when they are implemented as tangible user interfaces (TUIs) or when they serve as a background for the augmentation in handheld AR. This thesis includes four studies. In the first study, we used eye-tracking methods to look for evidences of the benefits associated to TUIs in the learning context. We designed a 3D modelling task and compared users' performance when they completed it using a TUI or a GUI. The gaze measures that we analysed further confirmed the positive impact that TUIs can have on the learners' experience and enforced the empirical basis for their adoption in learning applications. The second study evaluated whether the physical interaction with models of carpentry structures could lead to a better understanding of statics principles. Apprentices engaged in a learning activity in which they could manipulate physical models that were mechanically augmented, allowing for exploring how structures react to external loads. The analysis of apprentices' performance and their gaze behaviors highlighted the absence of clear advantages in exploring statics through manipulation. This study also showed that the manipulation might prevent students from noticing aspects relevant for solving statics problems. From the second study we obtained guidelines to design StaticAR which implements the magic-lens metaphor: a tablet augments a small-scale structure with information about its structural behavior. The structure is only a background for the augmentation and its manipulation does not trigger any function, so in the third study we asked to what extent it was important to have it. We rephrased this question to whether users would look directly at the structure instead of seeing it only through a tablet. Our findings suggested that a shift of attention from the screen to the physical object (a structure in our case) might occur in order to sustain users' spatial orientation when they change positions. In addition, the properties of the gaze shift (e.g. duration) could depend on the features of the task (e.g. difficulty) and of the setup (e.g. stability of the augmentation). The focus of our last study was the digital representation of the forces that act in a loaded structure. From the second study we observed that the physical manipulation failed to help apprentices understanding the way the forces interact with each other. To overcome this issue, our solution was to combine an intuitive representation (springs) with a slightly more formal one (arrows) which would show both the nature of the forces and the interaction between them. In this study apprentices used the two representations to collaboratively solve statics problems. Even though apprentices had difficulties in interpreting the two representations, there were cases in which they gained a correct intuition of statics principles from them. In this thesis, besides describing the designed system and the studies, implications for future directions are discussed.