I'm wondering what are the approaches for specifying goals for a trained deep-RL in deployment? E.g. how to tell a car drive agent to go to location $y$?
To elaborate, I understand that, for example, how Deep-RL can be used for solving an Atari games. But in these game, the goal is mostly fixed for an agent and hence can be encoded into the reward. But for something like a car driving, the goal is different if the destinations are different locations, e.g., sometimes I want to go to location $y_1$, sometimes $y_2$, etc. So it doesn't seem straightforward that I can encode the destination variables into the reward signal.
One solution is perhaps to add the destination location to the input state. But is there any other ways? Or maybe what are some of the approaches that doesn't rely on solely an RL component.