Continual Reinforcement Learning via Autoencoder-Driven Task and New Environment Recognition