Doubt in RL quiz

#1

Consider the situation of an atari tennis game. A game agent trained by your model competes against the computer’s game agent. The virtual player can move in the following directions- [north, south, east, west, north-west, north-east, south-west, south-east]. In this environment, the observation is an RGB image of the screen, which is an array of shape (210, 160, 3).

Which of the following approaches do you think is best suited to give an optimum result?

Capturing the screen at a particular time frame ⇒ Passing it to the model ⇒ Generating Reward

Capturing the screen at consecutive time frames ⇒ Stacking them together ⇒ Passing it to the model ⇒ Committing the predicted action ⇒ Generating Reward

None of the above

Capturing the screen at a particular time frame ⇒ Passing it to the model ⇒ Generating Reward

Capturing the screen at a particular time frame ⇒ Passing it to the model ⇒ Committing the predicted action ⇒ Generating Reward

#2

Hey @debjanihome, “Capturing the screen at consecutive time frames ⇒ Stacking them together ⇒ Passing it to the model ⇒ Committing the predicted action ⇒ Generating Reward”, will be the answer. The reason, being that the current action would depend on the series of past actions performed previously, you can imagine it as like lstm and rnn. Hence this is the best suited answer.

Hope this cleared your doubt.
Plz mark the doubt as resolved in my doubts section. :blush: