A Policy Adaptation Method for Implicit Multitask Reinforcement Learning Problems