My research equips robots to generate safer, more efficient movements and to learn motion patterns directly from human behaviour. My thesis research developed methods for robots to learn generalisable skills from human demonstrations and introduced a faster numerical integrator that enables efficient generation of trajectories from learned behaviour. I introduced a reactive motion generation framework capable of efficiently finding globally optimal trajectories while adapting to moving obstacles, and learning from past experiences. Additionally, I developed methods to speed up motion planning. My postdoctoral research further advanced this area by introducing approaches for robots to learn novel skills, spatial connections, and periodic motions, from human sketches.