Robot learning by observing humans activities and modeling failures

People needs are varied and different. Service robotics aims to help people to satisfy these needs, but not all the feasible programs can be preloaded into a robot. The robot have to learn new tasks depending on the circumstances. A solution to this challenge could be Robot Learning from Demonstration (RLfD). In this paper, a RLfD framework is described in its entire pipeline. The data are acquired from a low cost RGB-D sensor, so the user can act naturally with no need of additional hardware. The information are subsequently elaborated to adapt to the robot structure and modeled to overcome the differences between human and robot. Experiments are performed using input data coming from a publicly available dataset of human actions, and a humanoid robot, an Aldebaran NAO, is shown to successfully replicate an action based on human demonstrations and some further trials automatically generated from the learned model.