i probably made some embarassing mistake, but as i cannot find it, i ask you to have a look at this method.
It should calculate the angle between two vectors. And vectors are instances of Point (perhaps here is already a misconception?).
It is used in my flocking simulation PharoBoids (http://smalltalkhub.com/#!/~MartinWalk/Boids)
Here is my version of the method:
angleBetween: vector1 and: vector2 onError: aBlock
| cosinusOfAngle innerProductOfVectors productOfVectorsLengths|
innerProductOfVectors := (vector1 dotProduct: vector2).
productOfVectorsLengths := (vector1 r) * (vector2 r).
productOfVectorsLengths = 0 ifTrue: [ ^ aBlock value ].
cosinusOfAngle := innerProductOfVectors / productOfVectorsLengths.
^ cosinusOfAngle arcCos
But my Boids behave wrong when i use it. I found another implementation of the same problem by Igor Stasenko which works and looks like this:
angleBetween: p1 and: p2 ifDegenerate: aBlock
" Calculate an angle (in radians) between two vectors.
Evaluate a block, in case if calculation not possible because one of the vectors has zero length "
| x1 y1 x2 y2 dot2 n2 |
x1 := p1 x.
y1 := p1 y.
x2 := p2 x.
y2 := p2 y.
dot2 := x1 * x2 + (y1 * y2).
dot2 := dot2 * dot2.
n2 := (x1*x1 + (y1*y1)) * (x2*x2 + (y2*y2)).
n2 = 0 ifTrue: [ ^ aBlock value ].
^ (dot2 / n2) arcCos
Can anybody explain the problem of my method? M.
cos = (a dot b) / |ab|
while you are using
cos = (a dot b)^2 / (|ab|^2)
Rounding error?This may be the reason, igor's version is working with
2014-03-16 17:01 GMT+01:00 MartinW <[hidden email]>:
In reply to this post by MartinW
You can also explore the polar coordinates...
((p1 normalized) dotProduct: (p2 normalized)) arcCos ;-)
But maybe you just need
First of all, you could simplify your code with dotProduct, writing
dot2 := p1 dotProduct: p2.
On Sun, Mar 16, 2014 at 8:01 PM, MartinW <[hidden email]> wrote:
This was of course wrong, I don't know what theNow the question is, why do you think your method is wrong / the boids behave wrong?
other method computes, but not the angle between two vectors.
2014-03-16 18:41 GMT+01:00 Nicolai Hess <[hidden email]>:
Another observation in
ownDirection := self velocity - self position.
the own direction should be the same as the velocity.
Consider a Boid at position 100@0 with velocity 100@0 would give
ownDirection = 0@0.
But of course, the steering direction is still 1@0.
2014-03-17 10:07 GMT+01:00 Nicolai Hess <[hidden email]>:
> ownDirection := self velocity - self position.
Taking that purely at face value, it doesn't look right mixing distance(m) & velocity(m.s^-1) in a calculation. I guess you are integrating velocity over one unit of time...
(m.s^-1)*(s) - (m)
= (m) - (m)
Nicolai Hess wrote:
Yes i also wrote a test and i think my method is right.
When i use my method to calculate the field of vision the Boids always fly from top left to down right and if i turn on the circling option, they crowd in the bottom right quadrant.
If i give them a 360 degree field of vision by always returning true for isInFieldOfVision they behave correctly. This means they fly in a nice circle around the center and if circling is turned off, the swarm flies in different directions each time you start the simulation.
Perhaps the problem is not the calculation of the angle but is in the isInFieldOfVision: method...
2014-03-19 8:49 GMT+01:00 MartinW <[hidden email]>:
Nicolai Hess wrote
:) yes and I think the testcases in testIsInFieldOfVision
are wrong, or I misunderstand how the boids are moving.
subject has velocity 100@0, I understand this as "moves from left to right".
testobject other2 and other3 are above and below subect, and not behind and
in front, both should be visible with the fieldOfView.
Btw. the field of view is rather big, 0.75*Float Pi is 135 degree, but clock *and* counterclockwise
-> 270 degree field of view.
And it is funny to change the field of view to 0, after the boids started.
In reply to this post by MartinW
On 16 March 2014 17:01, MartinW <[hidden email]> wrote:
But my Boids behave wrong when i use it. I found another implementation of
yours looks correct.i'm just using squares to avoid taking square roots of vector lengths.
Oh, thank you. I made a new testcase. Which is unfortunately passing :( So i am still looking for a solution…
That's correct. Most birds have the eyes on the sides of their heads - thus they have a wide visual field. Some as wide as 360 degrees.
|Free forum by Nabble||Edit this page|