How do value-based methods like Deep Q-Networks handle continuous action spaces, and what limitations do these approaches have in real-world control?
