Ray reflection inaccuracy [SOLVED]



  • On 28/01/2015 at 04:44, xxxxxxxx wrote:

    Hi
    I'm in the middle of writing some code to calculate the ray reflections around tight corners inside a geometric object.  Sometimes the reflected ray bounces outside the object because it misses a close adjoining corner.  Anyway, do do this I need to manually test along the path of the reflected ray to see if it meets a new collision.  The code I'm using below is working toward this - but there's a general inaccuracy of the final position that ReflectRay calculates and the same position I calculate by taking the collision point and extrapolating along it.

    I'm pretty sure my math and code are correct, but my value (mypos)  and ReflectRay's value (pos) are very slightly different and it's bugging me - for example,

    Vector(-95.939, -75.666, -19.324) Vector(-97.445, -75.666, -19.324)

    global pos, vel

    predictedPos = pos + vel

    sourceobj = doc.SearchObject("Null") # The object the ray will start from 
        targetobj = doc.SearchObject("Cube") # The object the ray collides with    
        ray = GeRayCollider()                # Create a new GeRayCollider object
        ray.Init(targetobj, True)            # Assign the object to a variable

    start = pos
        direction = vel.GetNormalized()    #The direction the ray points.

    raylength = 1000

    CollisionState = ray.Intersect(start, direction, raylength)

    if CollisionState:
            hitpos = ray.GetIntersection(0)["hitpos"]
            collDist = c4d.Vector(hitpos - pos).GetLength()
            norm = ray.GetIntersection(0)["f_normal"]
            distance = ray.GetIntersection(0)["distance"]

    predDist = c4d.Vector(predictedPos - pos).GetLength()

    if collDist < predDist:
                norm.Normalize()
                reflect = ReflectRay(vel,norm) 
                mypos = hitpos + ( reflect.GetNormalized() * (vel.GetLength() - distance) )

    vel = reflect
                pos = pos + vel

    print pos, mypos

    else:
                pos = pos + vel
        else:
            pos = pos + vel



  • On 28/01/2015 at 06:31, xxxxxxxx wrote:

    Hi Glenn,

    can you perhaps give me a little more information. The two vectors you provide upfront are to be used as input data for your code example? In which way? And what is the expected result?
    Without having done any testing, I have a feeling in my guts, that you might have run into problems with floating point arithmetic. So while answering my questions, you may also want to have a look at the article about Floating-Point Weirdness on our PluginCafe blog.



  • On 28/01/2015 at 06:51, xxxxxxxx wrote:

    Hi Andreas,
    The two vectors I listed are the results of 'print pos, mypos' in the code.  They should be exactly the same but either one of x,y,z is always slightly off.  I suspected a floating point issue myself. If that's the case that's fine. I was just worried my calculations were wrong.



  • On 28/01/2015 at 17:22, xxxxxxxx wrote:

    Hi Glenn,

    checking your code it seems not to be a floating point issue.
    I can´t see exactly why you are doing this, but only adding the position to the whole reflected ray can´t give you the same result as there is no information of where the ray hits the surface.
    With utils.ReflectRay you´ll get the direction and the intensity of the ray just reflected.
    Your mypos calculation is right, a normalized reflection vector multiplied by the distance difference and added to the hitposition.

    The problems you have with the ray collider might come from just checking the first hit, which also can be a point or an edge and won´t have the normal you might need for your calculation. Just a guess...

    Best wishes
    Martin



  • On 29/01/2015 at 00:46, xxxxxxxx wrote:

    Thanks Martin,
    The thing is even when my object hits a perfectly flat surface - it still shows a slight discrepancy - it shows every collision no matter where in fact.

    Anyway, I cobbled this code together - it's a recursive call which takes an incoming position and velocity, and tries to work out all collisions.  From looking at it it seems to work - catching small tight corners etc.  It also doesn't make use of my manual calculations, but uses ReflectRay entirely.  (I've left in the debug code which compares my mypos error though)

    If you spot any errors in this - I'd be grateful to learn, thank you.

    def getAllCollisions(pos,vel2,count) :
        # print vel
        predictedPos = pos + vel2

    # sourceobj = doc.SearchObject("Null") # The object the ray will start from 
        targetobj = doc.SearchObject("Cube") # The object the ray collides with    
        ray = GeRayCollider()                # Create a new GeRayCollider object
        ray.Init(targetobj, True)            # Assign the object to a variable  
        start = pos
        direction = vel2.GetNormalized()    #The direction the ray points.

    raylength = 1000

    CollisionState = ray.Intersect(start, direction, raylength)

    if CollisionState:
            print '........................'
            hitpos = ray.GetIntersection(0)["hitpos"]
            norm = ray.GetIntersection(0)["f_normal"]
            distance = ray.GetIntersection(0)["distance"]

    if distance < vel2.GetLength() :
                norm.Normalize()
                reflect = ReflectRay(vel2,norm)

    mypos = hitpos + ( reflect.GetNormalized() * (vel2.GetLength() - distance) )
                print mypos 
                print pos + reflect # correct value ?

    vel2 = reflect

    count += 1
                if count < 10:
                    vel2 = getAllCollisions(pos,vel2,count)
        return vel2



  • On 29/01/2015 at 06:42, xxxxxxxx wrote:

    Hi Glenn,

    you again assume that ReflectRay will give you the right position without feeding the necessary hitposition into the equation.
    I attached a small snippet which deals with all collisions you can get from a given velocity, position and a bounce limit.
    The precision value is needed otherwise you will end up not to know if your are inside or outside the volume and therefore heading in the wrong direction, as you "stick on the surface".

    The code is heavily commented I hope this helps?
    Best wishes
    Martin

      
    import c4d  
    from c4d import utils  
      
      
    def getAllCollisions(pos,vel,count,bounceList) :  
        
      print pos, "the start position"  
      print vel, "the velocity vector"  
      print count, "the bounce count limit to ten"  
      precision = 0.000001 #distance tolerance to the surface avoiding unwanted catch at the surface      
            
      # initialize the ray collider with the target  
      targetobj = doc.SearchObject("Cube") # The object the ray collides with      
      ray = utils.GeRayCollider()          # Create a new GeRayCollider object  
      ray.Init(targetobj, True)            # Assign the object to a variable    
        
        
      # velocity values  
      direction = vel.GetNormalized()    #The direction the ray points.   
      intensity = vel.GetLength()        #The constant speed(intensitiy) the ray traverses  
      print direction, "direction"     
      print intensity, "the new raylength"  
        
        
      # raylength is the velocity intensity and will decrease the longer the ray is on his way when reflecting.  
      raylength = intensity  
      
      #check collisions  
      CollisionState = ray.Intersect(pos, direction, raylength)  
      
      
      if CollisionState and count < 10:  
          print '........................'  
            
          #take the nearest collision  
          hitpos = ray.GetNearestIntersection()["hitpos"]  
            
          norm = ray.GetNearestIntersection()["f_normal"]  
          norm.Normalize()   
      
          distance = ray.GetNearestIntersection()["distance"]          
            
          #the normalized reflection vector  
          reflect = utils.ReflectRay(direction,norm)  
      
          #the collision point refined by the precision value to avoid catch at surface given in targetobj local space  
          hitpos = (hitpos + precision*reflect)  
          print hitpos,"the collision point"  
             
          mypos = hitpos + ( reflect * (intensity - distance) )  
          print mypos ,"the new calculated endposition------------"  
      
          vel2 =  mypos-hitpos   
          print vel2 ,"the new velocity vector"  
            
          #the new calculated position and velocity for the next possible bounce          
          bounceList.append([hitpos,vel2])  
          count += 1  
      
          #recursivly searching for more bounces  
          getAllCollisions(hitpos,vel2,count,bounceList)  
                    
          return bounceList  
      
      
      else:   
            
          return []  
           
      
    def main() :  
        
      sourceobj = doc.SearchObject("Null")  
      
      pos = sourceobj.GetAbsPos()  
      vel = c4d.Vector(0,0,800)  
      count = 0  
        
        
      bounceList = [] #store all collision points with their velocities  
      print getAllCollisions(pos,vel,count,bounceList)  
        
        
        
        
      
    if __name__=='__main__':  
      main()  
      
      
    


  • On 29/01/2015 at 09:03, xxxxxxxx wrote:

    Hi Martin,
    Thanks so much for this - the code works perfectly and is well explained.
    I see what you mean about ReflectRay and my assumption of a hitpoint.

    Anyway - in my main loop (running every frame), if there has been any collisions then I update my new pos and vel like so

    global pos, vel
        bounceList = [] #store all collision points with their velocities
        getAllCollisions(pos,vel,0,bounceList)

    if len(bounceList) > 0:
            newpos = bounceList[-1][0]
            bounceVel = bounceList[-1][1]
            pos = newpos + bounceVel
            vel = vel.GetLength() * bounceVel.GetNormalized()

    else:
            pos = pos + vel

    Incidentally, the reason I shied away from using raylength for testing is because I saw this
    https://plugincafe.maxon.net/topic/5785/5839_geraycollider

    there seems to be problems if it is less than 100, but I haven't noticed anything so far.

    Thanks again!
    Glenn.



  • On 29/01/2015 at 11:18, xxxxxxxx wrote:

    Hi Glenn,

    I´m glad this helped.

    For your task there should´nt be any problem with the raylength.
    Another approach for other tasks could be to compare every collision in one "shot" and delete the double ones.

    It´s important to know that ray collider works in local space of the collision object.

    regarding your main loop:
    I assume that you have precalculated your velocity vector in the dimension of time.
    your vel is the way the ray travese in one time unit.
    In this case the next position for let´s say the next frame, will be the last mypos calculated in the getAll collisions function multiplied by the object matrix.
    The new vel will be, assuming we have a constant speed with no acceleration, the last reflect vector multiplied by the intensity we calculated at the beginning, as the speed doesn´t change.

    Best wishes
    Martin



  • On 29/01/2015 at 11:43, xxxxxxxx wrote:

    in other words

      
    def main() :  
        
      sourceobj = doc.SearchObject("Null")  
      
      pos = sourceobj.GetAbsPos()  
      vel = c4d.Vector(0,0,800) #initial vel  
      count = 0  
            
      bounceList = [] #store all collision points with their velocities  
      print getAllCollisions(pos,vel,count,bounceList)  
        
      targetobj = doc.SearchObject("Cube")  
      matr = targetobj.GetMg()  
        
      lastHitPos = bounceList[-1][0]  
      lastVel = bounceList[-1][1]  
      intensity = vel.GetLength()     
        
      newVel = lastVel.GetNormalized()*intensity      
      newPos = (lastHitPos + lastVel)*matr  
        
      print newPos,newVel  
      
      
    if __name__=='__main__':  
      main()  
    


  • On 30/01/2015 at 06:47, xxxxxxxx wrote:

    Hi Martin
    In my test scene I just have a null with a small object as its child - the null bounces around inside the cube at a constant velocity.  And everything looks to be working perfectly.  It's useful to know about multiplying the object matrix though.  I'll probably need to do that in the future.
    Thanks again,
    Glenn.



  • On 01/02/2015 at 04:59, xxxxxxxx wrote:

    Hi Glenn,

    great!
    I can almost imagine your small objects bouncing around(funny sentence while typing it but wasn´t meant to be offensive in any way😉)

    Referring to Andreas' feeling in his guts (nice article by the way) and for the sake of completeness:

    1.There is a problem with the floating point precision while "sticking" at the surface, but this will be the case even if we have much more precision in computer technology.

    2.The starting point needs to be within the object's space, too, if you move the object from world origin.

    3.If you use multiple rays in your code but with only one static object, the ray init function must be called only once, as it seems just preparing the object for collisions.
    This becomes obvious once your collision object has a looooot polygons.

    Curious about what you´ll end up with!
    Martin



  • On 02/02/2015 at 09:49, xxxxxxxx wrote:

    Hi Martin,
    My bouncy objects I'm using to create traced lines inside geometry to create interesting abstract forms.  Below is a simple cube.
    One little thing you might know about - The lines don't update in realtime inside the editor view when I play the timeline.  The code is inside a generator and updates every frame.  It's a bunch of growing loft objects basically.  When I move the camera around I get a brief glimpse of what's happening, but that's all.  Ive tried playing with the following..

    c4d.EventAdd()
    c4d.DrawViews( c4d.DRAWFLAGS_FORCEFULLREDRAW)

    but no joy.. not the end of the world, I'm just glad the end result is working.



  • On 03/02/2015 at 01:12, xxxxxxxx wrote:

    Hi Glenn,

    nice your tracer kind of thing!
    If you want to do some further development I´ll sent you a pm.
    Best wishes
    Martin



  • On 03/02/2015 at 04:52, xxxxxxxx wrote:

    Thanks for the message Martin, will drop you a line soon!


Log in to reply