$2$ c$)$ Extra Credit: Make your algorithm even better

Prof. Everett's simple pure pursuit implementation used an average number of steps of $79\mathrm{.}44$ on the validation set. You can earn $1$ extra credit

point if you upload a second implementation to Gradescope achieves below $80$ steps on average $($while also achieving $100\%$ success$).$ You can

earn additional extra credit for having the best implementation in the class according to the Gradescope leaderboard.

You can try different strategies than just pure pursuit. Whichever control strategy you end up using, please use the BetterCont rol ler class

below for this part, which has the same class$/$function signatures as before.

Note: there will be a test suite that is similar but not identical to the validation method above to ensure your algorithm doesn't overfit to specific

cases.

$[]$ class BetterController:

def$\_\_$init$\_($self$,$ L$=3.)$:

$""\text{'}"$Store any hyperparameters here."$\text{'}"\text{'}"$

self.L $=$ L

def get$\_$action$($self$,$ obs: np$.$ndarray, goal: np$.$ndarray$)$ np$.$ndarray:

$""\text{'}"$Your implementation goes here""$\text{'}"$

raise NotImplementedError$()$

return np$.$array$([$linear$\_$speed, angular$\_$speed$])$