Update heightfield collision dynamically?

[Aphex]

Hi,

I need to update a heightfield collision body in realtime - is there a way to just change the height of single points on one quickly? Re-creating it completely is too slow for realtime.
I reckon this would be a useful feature for heightmaps if not already available

[ed] hmm benchmark fun... actually something else was eating up the cpu... but it would still be nice to be able to tweak a height level separately, for deformable terrain

[Leadwerks]

There is a structure that contains a pointer to the height data. I do this all the time.

[Aphex]

Hmm... Is that using the NewtonCollisionGetInfo fn?

[Aphex]

No that doesn't work... any clue to what this structure is?

[Julio Jerez]

NewtonCollisionGetInfo should do it.
It give yo uall teh infomation in the shape.

[Aphex]

When I modify the heightmap data pointed at in the info structure Visual studio tells me the stack has been corrupted - doesn't sound good..?

[Julio Jerez]

you call

NewtonCollisionInfoRecord collisionInfo;
void NewtonCollisionGetInfo collision, NewtonCollisionInfoRecord* collisionInfo);

the you must check that collision type is an actual SERIALIZE_ID_HEIGHTFIELD typr

if it is the member collisionInfo.NewtonHeightFieldCollisionParam;
containg the pointer to the elevation and everything else

Code: Select all

struct NewtonHeightFieldCollisionParam
{
   int m_width;
   int m_height;
   int m_gridsDiagonals;
   dFloat m_horizonalScale;
   dFloat m_verticalScale;
   unsigned short *m_elevation;
   char *m_atributes;
};

[Aphex]

OK here's my code:

Code: Select all

         NewtonCollisionInfoRecord collisionInfo;
         NewtonCollisionGetInfo(NewtonBodyGetCollision(mPhysicsBody), &collisionInfo);
         _ASSERT(collisionInfo.m_collisionType == SERIALIZE_ID_HEIGHFIELD);


When I run this (apparently benign) code I get this error on function exit:

'Run-Time Check Failure #2 - Stack around the variable 'collisionInfo' was corrupted.'

i.e. it thinks Newton is writing past the end of the info buffer??


[ed] If I add some padding to the Newton structure after m_atributes, the error goes away...

Code: Select all

   char *m_atributes;
   char buff[129*129];  // my heightmap = 129x129


Maybe Newton's copy of the attributes data is being allocated to just [width], not [width*height] or something similiar?

[Julio Jerez]

not it does no copy the arry o elevation ther may be a bug with size,
the side of NewtonCollisionInfoRecord is 128 byte

Are you are using double? maybe there is problem with data size there
please try this

char *m_atributes;
char buff[1024];

NewtonCollisionInfoRecord* collisionInfo = buff;
NewtonCollisionGetInfo(NewtonBodyGetCollision(mPhysicsBody), collisionInfo);

this one of teh reason I do no liek using data strutore in cross platform code

[Julio Jerez]

I just pasted thsi code in teh test demo, and I runnint in 32 and 64 bit, it seem to work find
NewtonCollisionGetInfo(NewtonBodyGetCollision(m_level), &collisionInfo);
_ASSERT(collisionInfo.m_collisionType == SERIALIZE_ID_HEIGHTFIELD);


this is the internal code tha support that funtion.

Code: Select all

void dgCollisionHeightField::GetCollisionInfo(dgCollisionInfo* info) const
{
   dgCollision::GetCollisionInfo(info);

   info->m_offsetMatrix = GetOffsetMatrix();
   info->m_collisionType = m_collsionId;

   dgCollisionInfo::dgHeightMapCollisionData& data = info->m_heightFieldCollision;
   data.m_width = m_width;
   data.m_height = m_height;
   data.m_gridsDiagonals = m_diagonalMode;
   data.m_verticalScale = m_verticalScale;
   data.m_horizonalScale = m_horizontalScale;
   data.m_atributes = m_atributeMap;
   data.m_elevation = m_elevationMap;
}


// Name: NewtonCollisionGetInfo
// Get creation parameters for this collision objects.
//
// Parameters:
// *const NewtonCollision* collision - is the pointer to a convex collision primitive.
// *NewtonCollisionInfoRecord* *collisionInfo - pointer to a collision information record.
//
// Remarks: This function can be used by the application for writing file format and for serialization.
//
// See also: NewtonCollisionGetInfo, NewtonCollisionSerialize 
void NewtonCollisionGetInfo(const NewtonCollision* collision, NewtonCollisionInfoRecord* collisionInfo)
{
   dgCollision *coll;
   coll = (dgCollision *)collision;

   TRACE_FUNTION(__FUNCTION__);

   _ASSERTE (sizeof (dgCollisionInfo) <= sizeof (NewtonCollisionInfoRecord));
   coll->GetCollisionInfo ((dgCollisionInfo*) collisionInfo);
}

[Aphex]

Yeah I'm using the double version. sizeof(NewtonCollisionInfoRecord) gives me 1208.
Setting the buff to 1024 prevents the error too.

[Aphex]

Also wondering if this is related to the strange behaviour I get on heightfields with poly winding of '1'...

[Julio Jerez]

and what would that be?
Newon only wind the faces on way.

you can set the orientation of the quads diagonals at creation time, but you can not change then at run time.

[Aphex]

Whoops yes I meant the diagonal orientation, not face winding
That problem is that my multibody vehicle wheels can sink into the heightmap body when the diagonals are set to '1'.

[Aphex]

Here's a wmv of it in action: http://homepage.ntlworld.com/martingbell/files/NewtonHeightmapProb01.wmv

[ed] hmm... in a bit of recursive thinking, maybe this problem was fixed by that raytrace bug you fixed a while back... but I haven't been able to use the newest sdk due to the vehicle problems