Boosting Exploration in Multi-Task Reinforcement Learning using Adversarial Networks