Reinforcement Learning for a Dexterous Manipulation Task