In this paper, the study and implementation of a task generation system that uses the information obtained from the user and already known cases is presented. One of the main objectives of the system is to introduce a new approach in robotics that takes into account the physical limitation of teaching and learning time, and thus the amount of knowledge that a robot can obtain of a given environment (tasks, objects, user preferences…), as a critical bottleneck of any robotic system. For this, the study of the Case Based Reasoning (CBR) problem is presented. Additionally, Base Trajectory Combination (BATC), a novel trajectory combination method based on a simplified CBR structure, using trajectories instead of high-level tasks, is proposed and explained. Finally, this system is tested with Moveit! as the simulation environment, using the humanoid robot TEO from Universidad Carlos III de Madrid as the robotic platform. The results of these experiments are also presented with the corresponding conclusions and future research lines.