My physical object won't hold still.

I'm making a little follower pet prim thing. And it's working fine. Except for one thing. Any time it hits an avatar or obstruction, it spins and rotates on all the axis' like mad. Makes it impossible to design any kind of top, bottom, front, or back... This is the script I'm using.

Can anyone help me pinpoint exactly how to keep it from spinning wildly?

vector offset = < -1, 0.5, 1.5>; //1 meter behind and 1 meter above owner's center.

default
{
state_entry()
{
llSetStatus(STATUS_PHYSICS, TRUE);
// Little pause to allow server to make potentially large linked object physical.
llSleep(0.1);
// Look for owner within 20 metres in 360 degree arc every 1 seconds.
llSensorRepeat("", NULL_KEY, AGENT, 20.0, PI,1.0);

}
sensor(integer total_number)
{ // Owner detected...
// Get position and rotation
vector pos = llDetectedPos(0);
rotation rot = llDetectedRot(0);
// Offset back one metre in X and up one metre in Z based on world coordinates.
// use whatever offset you want.
vector worldOffset = offset;
// Offset relative to owner needs a quaternion.
vector avOffset = offset * rot;

pos += avOffset; // use the one you want, world or relative to AV.

llMoveToTarget(pos,0.4); 
}
}

I understand that using "llLookAt" would probably do the trick, but I can't get it to work with this particular script.

Your question was answered less than a week ago in this forum:
/54/6e/324400/1.html

axis locking really is the best solution. but, if you can't lock the x and y axis as suggested in that thread (because you're allowing changes when going up a hill or some such), you can force the object back to near upright with a fun little function I've been playing with...

rotation uConformPlane2Z( rotation vRotRaw, float vFltStrength ){
	vector vVecFwd = <1.0, 0.0, 0.0> * vRotRaw;
	vector vVecTop = <0.0, 0.0, vFltStrength>;
	return llAxes2Rot( vVecFwd, vVecTop % vVecFwd,  vVecTop);
}

//--used in the following call
llSetRot( uConformPlane2Z( llGetRot(), 1 ) );

this will incrementally rotate the object closer to being level... the higher the strength, the bigger the snap amount. you may want to use it in conjunction with llRotTarget, and call it from not_at_rot

I have one complaint: The 3 vectors in llAxis2Rot are not mutually orthogonal except in the special case where vRotRaw == ZERO_ROTATION and in a few other special cases;
And vVecTop will only be a unit vector when vFltStrength == 1.0;
I don't know how much this means in your case but I have experienced cases where it was crucial!
"All three vectors must be mutually orthogonal unit vectors"
Reference: http://wiki.secondlife.com/wiki/LlAxes2Rot

If this sound familiar it is because I said it before:
/54/c6/322162/1.html
In that thread I offered a solution too

it should be (purposely) non orthogonal, for any rotation that can't be expressed as being solely around the z Axis... that's how it works... it averages that (based on the strength) to get a correction back towards the z-plane. I'm not up to writing a proof, or even a detailed process of the conformation, I can barely fight my way through quat math while making sense of it. but the action (and it's cause) are apparent enough...

someone better at the math of it could probably clean it up to fewer steps, but i'm not there yet =)

(and yes that's the thread it's from, but that formula wasn't actually a solution for the stated problem, instead it was an unexpected bonus for something else... like vulcanized rubber)

interesting side notes:
sqrt( 2 ) is the value at which it'll make a 50% correction from 90deg off the z plane (straight up or straight down).

also in those 2 cases alone (straight up/down), is the x facing forced towards +X regional.

I am learning so much, this is awesome. Thanks.