Latent Exploration for Reinforcement Learning