Fix head-pose prediction bug and increase time
The pose processor works in "tick" time units, so the velocities
passed to it need to be scaled accordingly.
Also increase prediction time to 50ms (for now) to make it more
noticeable.
Test: Manual verification.
Change-Id: I842049f497cfc1824ee33f0bcc85349916b20bc3
diff --git a/services/audiopolicy/service/SpatializerPoseController.cpp b/services/audiopolicy/service/SpatializerPoseController.cpp
index 440a7ff..a7bc31e 100644
--- a/services/audiopolicy/service/SpatializerPoseController.cpp
+++ b/services/audiopolicy/service/SpatializerPoseController.cpp
@@ -58,7 +58,7 @@
// twist (velocity). It should be set to a value that matches the characteristic durations of moving
// one's head. The higher we set this, the more latency we are able to reduce, but setting this too
// high will result in high prediction errors whenever the head accelerates (changes velocity).
-constexpr auto kPredictionDuration = 10ms;
+constexpr auto kPredictionDuration = 50ms;
// After losing this many consecutive samples from either sensor, we would treat the measurement as
// stale;
@@ -253,7 +253,8 @@
const std::optional<Twist3f>& twist, bool isNewReference) {
std::lock_guard lock(mMutex);
if (sensor == mHeadSensor) {
- mProcessor->setWorldToHeadPose(timestamp, pose, twist.value_or(Twist3f()));
+ mProcessor->setWorldToHeadPose(timestamp, pose,
+ twist.value_or(Twist3f()) / kTicksPerSecond);
if (isNewReference) {
mProcessor->recenter(true, false);
}