This paper studies the problem of evaluation of human 3d+t activities in Virtual Environments (VE) for Learning (VEL). Current evaluation methods focus mostly on: (i) the automatic identification of an ordered sequence of actions and/or (ii), an empirical analysis made by experts through the VE. In many cases, the learner's activity can be represented by some specific time series made of geometrical data of 3D artefacts. For the extraction and analysis of such Motions Of Interest (MOI), one can manually segment them among the traces, and/or use automatic approaches requiring a database of annotated examples. Both cases usually require too many resources to design such environments. Consequently, this work presents a method allowing teachers to quickly build, compare and evaluate a 3d+t learning activity in VE. This method is based on a semi-automatic approach combining the Dynamic Time Warping algorithm, with 3D reference shapes and few expert's demonstrations of the task to learn.