On the Fly Adaptation of Behavior Tree-Based Policies through Reinforcement Learning