Simple Recipe Works: Vision-Language-Action Models are Natural Continual Learners with Reinforcement Learning