Related with: Long-short term memory prediction of user's locomotion in Virtual Reality publication (Dataset)

Dataset: Captured motion data from 44 users.

Scenes:

SL -> Scene Lab.

SR -> Escape Room.

MF -> Shooter forest.

Since it is recorded inside a game engine and all records take place inside their processing, the timestamp is written down for each register (Time_sice_startup field). Additionally, the anonymized identification of the user is recorded (User field).

The dataset includes the following characteristics for Oculus Quest 2 HMD and each controller.

    DevicePosition (x, y, z): Position recorded.

    DeviceRotation (w, x, y, z): Rotation expressed with a quaternion.

    Forward (x, y, z): The unit vector that points to the specific device in the forward direction used in our new model. It can also be obtained by rotating $(0,0,1)$ with the quaternion.

    DeviceVelocity (x, y, z): Linear velocity of that device in that frame. It represents the rate of change in position.

    DeviceAcceleration (x, y, z): Linear acceleration of that device in that frame.

    DeviceAngularVelocity (x, y, z): The angular velocity vector in that frame of the device is measured in radians per second.

    DeviceAngularAcceleration (x, y, z): The angular acceleration at that frame. 

Also for each goal in the scene:

    GoalName (x, y, z): Position of that goal. If the element is static, the same position will always be recorded.

    GoalName_Quat (w, x, y, z): As in the previously defined fields, a rotation is expressed as a quaternion.

    GoalName_LocalScale (x, y, z): Scale of that element locally related to its parent in the hierarchy. They have no relatives in their hierarchy, so it is the real scale.