Learning and reusing primitive behaviours to improve Hindsight Experience Replay sample efficiency