# [Community Puzzle] Search Race

The checkpoints are in the order you get them first turn. Tests and validators also.

2 Likes

Ok thank you

3 Likes

Hi, I am trying to create a simulator myself for testing locally and am not able to match my outputs to the game outputs. Q: When the program is calculating the new position and velocity based on my inputs, are all evaluations done as integers? or they are first calculated as floats and then truncated/rounded?

For example, I took the test 1. Just to get the game data, I am pointing my car towards the 1st checkpoint and these are the outputs given by game

P=(10353,1986) V=(0,0) Ang=(161)
P=(10258,2019) V=(-80,28) Ang=(161)
P=(10083,2080) V=(-148,52) Ang=(161)
P=(9840,2165) V=(-205,72) Ang=(161) - not able to match Vx here
P=(9540,2270) V=(-254,89) Ang=(161)
P=(9191,2392) V=(-296,103) Ang=(161)

can anyone please explain how these are calculated in the game.

Check this post higher up: [Community Puzzle] Search Race - #27 by Illedan

1 Like

My feedback :
16800.78 with a heuristic, taking care about :

• the best angle to go to the second next checkpoint Replay : Tokyo drift (look at the very first checkpoint).
• to aim the second next checkpoint if the first is aligned Replay : round and round
• to adapt the thrust according to the angle of the first next checkpoint.

I think iāll try to do a real IA with a simulation, but you know you can do something correct with a heuristic.

6 Likes

Dāoh, my GA passes all tests and it passes all 68 tests of the demo version, including those with test * (Validator).
But i cannot get a 100% with a submit on the release version, it always fails 3-4 tests.

Whatās your time limit? For some optims it is be much less than 50ms, more like 35ms.

995ms for the first turn, then 45 for each subsequent loop after the first input arrived.
Timeouts have been my first suspect, but all tests even succeed with 1000/49.

Iāve tried conservative and optimistic settings, but always : all tests succeed, submit validators fail.

Is there a way i can see, why a certain validator has failed after a submit? To differentiate between e.g. code failure and a timeout?

Could someone please, give simple instructions how to set up the referee for local testing?

I donāt understand how the angle varible works. Itās not like in the Mad pod racing next_cp_angleā¦ Can someone explain to me a little bit about how this variable works?

Itās the angle of the car, no relation to checkpoint. 0 means it points right, 90 down, 180 left, 270 up, 45 its down right etc.

1 Like

I have a problem . Itās slow. Look at my replay. How can I improve ?
https://www.codingame.com/replay/767501374

Hi all, has anybody else tried implementing A*? I did implement A* for the path finding in the CSB years back (when it was a contest) and it worked really well for me. I donāt have that code anymore (neither does CG).
Now I tried re-implementing the same thing for this, but it just doesnāt seem to work. The main problem seems to be that I find it really difficult to nail down the the heuristic value for estimating the ānumber of steps to the goalā. Did anybody attempt A* and did it work and would you share your ideas for the heuristic with me?

Oh btw Iāve also been playing with some Bezier/Hermite curves/splines to plan the path (with some GA optimization) and then find out with some DFS how the curve can be followed, has anybody tried that?

This looks a fast as can be. What exactly would you like to improve?

The stability of the car.

Hey peeps ! Iām new to GA, so trying to build it from ground up, but for some reason I never even get to first checkpoint. Can someone help ?

Please avoid posting codes on the forum. You may describe your approach first, and if anyone wants to read your code, you can share with them via private message.

Hi @5DN1L How are you ? Can I share my code with you ?

Hi @franca1s Iām ok, thanks. Hope youāre well too. You can share your code with me, and Iāll try my best to see what goes wrong (no guarantee though ).

Edit: The issue was found: the simulations were too short to reach the first checkpoint.