The black box driver estimating the handler position and velocity using, it seems, a Kalmann and an inertial filters, obtains degraded results as compared what it obtained before.
More precisely, if the handler comes from outside the FOV of the tracking cameras and enters this FOV, the velocity is completely irrelevant (sometimes pointing rearward while the handler is moved forward for one or even several frames).
His is easy to reproduce by printing the angle between velocity and the player camera direction, and by moving (slowly or quickly) from behind the head to in front of the head.
Before throwing objects was a real pain in VR before (example : throwing objects in SuperHot VR). Either the object was falling on the floor, or the throwing direction was wrong up to 90°.
Now it is worse, since the error is random up to 180°.
You would have 2 ways of solving this :
1) Going back to the previous velocity and position engine, and stop modifying it (how can we expect a developed game to react consistently otherwise?)
2) Exposing the raw data (really raw, not already filtered), enabling us to do the job as we wantit to be done (I'm a scientist in image processing for 25 years)
Thank you @Eire_Max. I will ask them. Each time I see a driver update on the headset, I feel a cold sweat. My development push the Meta Quest over its limits, and any regression or modification can put my projects in jeopardy.
