Investigating Generalisation in Continuous Deep Reinforcement Learning