Project due Sep $4,202407$:$59$ EDT

In this section, you will implement the Q$-$learning algorithm, which is a model$-$free algorithm used to learn an optimal Q$-$function. In the tabular setting, the algorithm maintains the Q$-$value for all possible state$-$action pairs. Starting from a random Q$-$function, the agent continuously collects experiences and updates its Q$-$function.

From now on$,$ we will refer to as $$an action" although it really is an action with an object.

Q$-$learning Algorithm

The agent plays an action at state $,$ getting a reward and observing the next state $.$

Update the single Q$-$value corresponding to each such transition:

Tip: We recommend you implement all functions from this tab and the next one before submitting your code online. Make sure you achieve reasonable performance on the Home World game

Single step update

$1$ point possible $($graded$)$

Write a function tabular$\_$q$\_$learning that updates the single Q$-$value, given the transition date $.$

Reminder: You should implement this function locally first. You can read through the next tab to understand the context in which this function is called

Available Functions: You have access to the NumPy python library as np$.$ You should also use constants ALPHA and GAMMA in your code