Head offset

By default, nothing changes the location of the camera component apart from its attachment point (socket, bone, spring arm etc).

At least I can't see anything that does :)

It seems like a strange way to go about things, but I think I can make it work. I've been rewriting my project from scratch in 4.5.1 over the last 2 nights.

I'm not making any progress. As soon as I press alt+enter to put the window in the rift, my view detaches itself from its socket and starts doing whatever the Oculus plugin wants it to do...

I've commented out the bodies of UpdatePlayerViewPoint, ApplyHmdRotation, and UpdatePlayerCameraRotation, but the Rift still hijacks my view. If I could locate where the hijack was ocurring, I could modify it to do math for my animations and the camera in the same place, but comments in OculusRift.cpp are sparse to say the least.

EDIT: Ok, I found it. It seems to happen in CalculateStereoViewOffset;

//const FVector vHMDPosition = DeltaControlOrientation.RotateVector(CurHmdPosition);
//ViewLocation += vHMDPosition;

Commenting out those lines prevents the hijack.

Create your own player camera manager and in your version of
APlayerCameraManager::UpdateViewTarget(), change this:

	if (bFollowHmdOrientation)
	{
		if (GEngine->HMDDevice.IsValid() && GEngine->HMDDevice->IsHeadTrackingAllowed())
		{
			GEngine->HMDDevice->UpdatePlayerCameraRotation(this, OutVT.POV);
		}
	}

To this:

	if (bFollowHmdOrientation)
	{
		if (GEngine->HMDDevice.IsValid() && GEngine->HMDDevice->IsHeadTrackingAllowed())
		{
			FVector ViewForward, ViewRight, ViewUp;
			FRotationMatrix(OutVT.POV.Rotation).GetScaledAxes(ViewForward, ViewRight, ViewUp);
			
			GEngine->HMDDevice->UpdatePlayerCameraRotation(this, OutVT.POV);

			FQuat OrientationAsQuat;
			FVector Position(0.0f);

			GEngine->HMDDevice->GetCurrentOrientationAndPosition(OrientationAsQuat, Position);

			OutVT.POV.Location -= (Position.X * ViewForward);
			OutVT.POV.Location -= (Position.Y * ViewRight);
			OutVT.POV.Location -= (Position.Z * ViewUp);
		}
	}

And see what happens.

This is slightly different to my implementation, which uses a custom view rotation variable instead of controller rotation.

Just FYI: I haven't given up on this issue, I really need to find some time and think this through. Like I tried to explain earlier, there are a lot of variables here and one of the most important ones is latency. Therefore, a lot of stuff is happening 'automatically' (such as applying orientation/position to 'camera').

"artyom17" wrote:
The GetCurrentOrientationAndPosition function returns current orientation and position offset from the zero point.

How is this zero point determined?
It is just using the pawn/character eye height?
i.e. GetActorLocation() + FVector(0.0f, 0.0f, BaseEyeHeight);

This is described in our doc: http://static.oculus.com/sdk-downloads/documents/Oculus_Developer_Guide_0.4.3.pdf, chapter 7.1.1.
By default, the "Zero point" is the origin point that is located one meter away from the camera in the direction of the optical axis but with the same height as the camera. The default origin orientation is level with the ground with the
negative z axis pointing towards the camera. In other words, a headset yaw angle of zero corresponds to the
user looking towards the camera (please, refer to the doc chapter 7.1.1 for a pic).
The zero point location got corrected once ResetPosition() (or ResetOrientationAndPosition()) function is called; in this case, the current location of the headset becomes a 'zero point'.
An example. Let assume your head is located in the point of origin in the tracking camera frustum (again, see the doc, chapter 7.1.1). In this case GetCurrentOrientationAndPosition will return FVector(0,0,0). Now you shift you head to the right by 10 cm (assuming World2Meters scale is 100). The newly reported position will be FVector(0, 0.10*100, 0). Left 10 cm - FVector(0, -0.10*100, 0) (assuming there is no rotation; if there was rotation then the returned vector is rotated as well).
In the case, if you used ResetPosition / ResetOrientationAndPosition, the current HMD position (at the moment of the reset function call) is saved as BaseOffset and it is used to correct the returned vector by subtracting BaseOffset from the position to be returned (effectively bringing the 'zero point' to that location).

Ok.
A bit confused, as what I've just read and what is happening doesn't make sense.

If I understand correctly, if the I place my DK2 down on the table so it doesn turn or move, then load up the game, the only difference between the VR and non-VR view should be the offset for each eye position?

"KrisRedbeard" wrote:
Ok.
A bit confused, as what I've just read and what is happening doesn't make sense.

If I understand correctly, if the I place my DK2 down on the table so it doesn turn or move, then load up the game, the only difference between the VR and non-VR view should be the offset for each eye position?

Uhm....In the case if your headset was in field of view of the tracker camera then PosB (your have the headset picked up) minus PosA (the headset is laying on a table) will give you the offset. But the PosB and PosA themselves will represent offsets from the origin point, which is fixed to 1m from the tracker camera (see the doc, chapter 7.1.1).

Just go to FOculusRiftHMD::GetCurrentPose function (file Engine/Plugins/Runtime/OculusRift/Source/Private/OculusRiftHMD.cpp) and uncomment the following line:

UE_LOG(LogHMD, Log, TEXT("P: %.3f %.3f %.3f"), CurrentPosition.X, CurrentPosition.Y, CurrentPosition.Y);

Replace the 'CurrentPosition' by the 'CurrentHmdPosition', build and run (in VS Debugger). Check what is being printed in the Debug Output, while you are moving the headset in front of the tracker camera.

Note, I've corrected my previous reply: when you shift to the right you get Y change to plus, left - Y change to minus (previously I mistakenly used X -axis). Also used wrong units (should be 0.1 instead of 10 for 10 cm).

P.S. Seems like we need to document this moment well. Noted.