Jump to content

Rotational tracking

From VR & AR Wiki
See also: Tracking and Positional Tracking
The three rotational axes tracked in orientation tracking: pitch, yaw and roll.

Rotational tracking is a term used to describe how a piece of hardware determines how tilted something is. There are three different ways that an object can rotate, we call each of these ways a degrees of freedom, and you can track any number of degrees of freedom. Rotational tracking measures the three rotational degrees of freedom alone, so a device that tracks only rotation is described as having three degrees of freedom, or 3DoF.[1] The three rotational degrees of freedom are:

  • Pitch - Tilting backward and forward, like a plane ascending or descending
  • Yaw - Left and right movement, like a plane turning left or right while remaining level
  • Roll - Clockwise or counter-clockwise movement, like a plane banking to make a turn

In order to get an accurate idea of exactly how an object is oriented, all three types of rotation have to be measured. Traditionally this would be done using devices called IMUs implanted into an object.

Rotational tracking is a vital technology used not only in VR and AR HMDs but also in Input Devices.

Rotational axes

Any rigid object's orientation in three-dimensional space can be described by how much it is rotated about three perpendicular axes. The aviation-derived terms pitch, yaw and roll are the standard way to name these rotations.[1] Pitch is rotation about the lateral (side-to-side) axis, the motion of nodding the head "yes." Yaw is rotation about the vertical axis, the motion of shaking the head "no." Roll is rotation about the longitudinal (front-to-back) axis, the motion of tilting the head toward one shoulder.[2] Because these three values together fully define which way a head-mounted display or controller is facing, capturing all three is what allows a VR application to redraw the scene so that it stays fixed in the virtual world as the user looks around.

Sensors used

Rotational tracking is performed by an IMU (inertial measurement unit), a small package that combines several inertial sensors. The three sensor types most relevant to orientation tracking each measure a different physical quantity:[3]

Sensor Measures Strength Weakness
Gyroscope Angular velocity (rate of rotation) about each axis Fast, smooth response to motion; tracks all three of pitch, yaw and roll Readings drift over time as small integration errors accumulate
Accelerometer Proper (linear) acceleration, including the constant pull of gravity Gives an absolute reference for pitch and roll from the gravity vector Noisy; cannot sense yaw; gravity reference is swamped during fast motion
Magnetometer Strength and direction of Earth's magnetic field Gives an absolute heading (yaw) reference relative to magnetic north Easily distorted by nearby metal and electromagnetic interference

A unit that carries a three-axis gyroscope, a three-axis accelerometer and a three-axis magnetometer is often marketed as a "9DoF" or 9-axis IMU. This figure is the sum of the individual sensor channels (3 + 3 + 3) rather than nine real-world degrees of freedom; all of the sensors still resolve the same three rotational degrees of freedom.[2]

Sensor fusion and drift correction

No single inertial sensor produces a reliable orientation on its own, so a process called sensor fusion combines their outputs to exploit each sensor's strengths while cancelling its weaknesses.[3]

The gyroscope is the primary source of orientation: integrating its angular-velocity readings over time produces a running estimate of how the device has rotated. This works well over short intervals, but each integration step adds a little error, and a slowly changing sensor bias compounds the problem, so the estimate steadily drifts away from the true orientation.[4]

To stop this drift, the fusion algorithm pulls the estimate back toward absolute references that do not drift. When the device is relatively still, the accelerometer senses which way gravity points and corrects pitch and roll; the magnetometer senses magnetic north and corrects yaw, the axis the accelerometer cannot resolve.[3] A common way to blend these signals is the complementary filter, which trusts the gyroscope for fast, short-term changes and the accelerometer and magnetometer for slow, long-term accuracy, effectively high-pass filtering the gyroscope and low-pass filtering the absolute references.[4] More elaborate schemes such as the Kalman filter weight each measurement dynamically according to its estimated reliability.[3] Orientation is frequently held internally as a quaternion rather than as raw pitch, yaw and roll angles, because quaternions avoid the gimbal-lock problem of Euler angles and interpolate smoothly during fusion.[4]

Relationship to positional tracking

Rotational tracking should not be confused with positional tracking. The six degrees of freedom of a body in space split into two groups of three: the three rotational degrees of freedom (pitch, yaw, roll) and the three translational degrees of freedom (moving forward and back, up and down, left and right).[2] Rotational tracking covers only the first group. A device that adds translation tracking on top of rotation is said to have six degrees of freedom, or 6DoF.[1]

Because a purely rotational, 3DoF system knows only which way the user is facing and not where the user is, the wearer can look in any direction but cannot lean in to inspect an object, dodge sideways, or step around a corner.[2] Translation does not produce the parallax cue that helps the brain judge depth, and the mismatch between what the eyes see and what the inner ear feels can contribute to motion sickness, which is one reason 6DoF tracking became the standard for room-scale VR.[2] Adding positional tracking generally requires more than an IMU, for example external sensors or inside-out cameras whose observations are fused with the IMU data.[1]

Use in VR and AR

Pure 3DoF rotational tracking defined the first wave of modern consumer headsets. The original Oculus Rift DK1 tracked the rotation of the user's head with a sensor that combined a gyroscope, an accelerometer and a magnetometer, the three sensors of a 9-axis IMU, and was a 3DoF device.[5] Later mobile and standalone headsets that relied on rotational tracking only included Google Cardboard, the Samsung Gear VR, Google Daydream and the Oculus Go.[1] Many handheld VR controllers, including the small remotes that shipped with those early standalone headsets, are likewise 3DoF input devices that report only their orientation. Rotational tracking also remains a building block of 6DoF systems, where the IMU's fast orientation estimate is fused with positional data from cameras or external trackers.[1]

References