User talk:Dsekercioglu/MEA

From Robowiki
Jump to navigation Jump to search

Contents

Thread titleRepliesLast modified
And to make it even faster2215:08, 25 September 2017
Principles behind?101:56, 24 September 2017

And to make it even faster

       double x = bulletSpeed;
       double a = -3.508129323E-5;
       double b = 2.460363294E-3;
       double c = -6.666318894E-2;
       double d = 8.545020365E-1;
       double e = -3.337830707;
       double angle = a * x * x * x * x + b * x * x * x + c * x * x + d * x + e;
       return Math.sin(angle) / (bulletSpeed / 8 - Math.cos(angle));
Xor (talk)13:33, 23 September 2017

btw, you can try newton's method to find the max value ;)

Xor (talk)13:40, 23 September 2017

I know. But calculated numbers weren't exact and Wolfram Alpha's computation time exceeds.

Dsekercioglu (talk)13:43, 23 September 2017
 

Precision is up to 9 numbers. I think that it is enough.

Dsekercioglu (talk)13:45, 23 September 2017
 

Sorry, I didn't see that you changed the code. In my robot I did it like that but it is more understandable with Math.pow(x, y).

Dsekercioglu (talk)13:55, 23 September 2017

Math.pow makes me think something like e^x, but it turned out to be polynomial.

Xor (talk)13:59, 23 September 2017
 

Wow, I had no idea that the escape angle could be 15% bigger than traditional MEA calculation at the most common bulletspeed of 1.9 . . . Seems like my list of things to do for GresSuffurd to become top-10 again, gets bigger and bigger.

GrubbmGait (talk)16:56, 23 September 2017

this formula is bigger, simply because it is wrong

Xor (talk)01:44, 24 September 2017

Can you give the formula you used to prove that my formula is wrong?

Dsekercioglu (talk)10:08, 24 September 2017

I published the reason why it is wrong in another thread.

Xor (talk)10:49, 24 September 2017
 

I also agree that it's wrong. But why is it bigger than the correct one? Can you clarify that?

I understood that the problem with this is that it assumes that it's optimal to move in a fixed retreat angle (like moving in a perfect orbit), when it's actually true that the best retreat angle isn't a function only of the initial position, but it changes as you move around the wave. Is that the real issue?

Rsalesc (talk)12:51, 24 September 2017

I knew it's wrong because I had made the same mistake in my previous research.

It's bigger, simply because it uses some smaller distance to calculate the radians.

The correct one is:

<math>\theta = \int_0^{t_1} \frac{v_{lateral}}{d_{initial} + v_{retreat} t}dt</math>

where <math>\theta</math> is the escape angle, <math>t_1</math> is the total time (from bullet fired, to bullet hit). (this formula is true even if <math>v_{lateral}</math> or <math>v_{retreat}</math> is not constant)


Note that <math>d_{initial} + v_{retreat} t</math> is bigger than (his) <math>(v_{bullet} - \overline{v_{retreat}})t_1</math> almost all of the time,

as <math>d_{initial} =

(v_{bullet} - \overline{v_{retreat}})t_1</math>. 

Therefore his integral (<math>\theta = \int_0^{t_1} \frac{v_{lateral}}{(v_{bullet} - \overline{v_{retreat}})t_1}dt = \frac{v_{lateral}}{v_{bullet} - \overline{v_{retreat}}}</math>, assume <math>v_{lateral}</math> is constant) is bigger.

Xor (talk)15:02, 24 September 2017
 
  • Actually, I'm not sure that Traditional MEA is correct. It assumes that the bot doesn't change it's move angle until the wave hits. Because of that you can't get a MEA higher than Pi / 2 with Traditional MEA formula. When you move orbitally, lateralVelocity / (bulletSpeed + advancingVelocity) is the formula that will give you the EA so you can get a EA higher than Pi / 2.
Dsekercioglu (talk)16:09, 24 September 2017

You can't get escape angle higher than Pi/2 by the traditional formula simply because it is impossible.

If your formula can, it must be wrong.

Xor (talk)16:12, 24 September 2017
 
 
 
 
 

Principles behind?

in your calculation, you are calculating escape angle theta as theta = v_lateral / (v_bullet - v_retreat)

however, imo this formula is only true when you are moving orbital exactly ;)

Else, you have to calculate some integral, as the change rate of theta is changing everytime the distance changes. (Like my version)

Xor (talk)13:57, 23 September 2017

imo if you really want to calculate the max escape in a brute force way — try every possible paths a robot can move.

And that would be, for an estimated hit time of 50ticks, if you try 360 angles per tick — there will be 360^50 angles in total (~10^128). Anyway, you can always prune a lot of them.

Xor (talk)01:55, 24 September 2017