Adapting Virtual Embodiment Through Reinforcement Learning