This work presents a novel approach on task oriented control of a humanoid robot through the implementation of a cognitive architecture. The architecture developed here provides humanoid robots with systems that allow them to continuously learn new skills, adapt these skills to new contexts and robustly reproduce new behaviours in dynamical environments. This architecture can be thought of as a first stepping stone upon which to incrementally build more complex cognitive processes, providing this way a minimum degree of intelligence for the humanoid robot. Several experiments are conducted to prove the validity of the system and to test the operation of the architecture.