Skip to content

Latest commit

 

History

History
60 lines (54 loc) · 6.2 KB

skeleton_auto_config.md

File metadata and controls

60 lines (54 loc) · 6.2 KB
Raw

How to use Skeleton Auto-Configuration

Before using skeleton auto-configuration, you must "reset" your body proportion values by standing straight up and pressing the "Reset All" button under the "Body proportions" section. If this is not done, then the height value used in calculations will be incorrect.

VERY IMPORTANT: During recording, you must keep your heels in the same position, otherwise the values will be invalid.

To use skeleton auto-configuration, follow these steps:

  1. Stand straight up and pressing the "Reset All" button under the "Body proportions" section of the SlimeVR server window
  2. Locate and press the "Auto" button under the "Body proportions" section to open the skeleton auto-configuration window
  3. Stand in front of a chair
  4. Keep your heels on the ground in the same position for the duration of recording
  5. Press the "Start Recording" on the skeleton auto-configuration window, the text on the button will change to indicate that recording has started - recording will take approximately 20 seconds
  6. Move until the text on the button changes back to "Start Recording", the current best known movements for calibration are as follows:
    1. Twist upper body left, then lean right (toward toes)
    2. Twist upper body right then lean left (toward toes)
    3. Sit down on a chair then stand up
    4. Bend down to touch toes then back up
    5. Sit down and wiggle knees then stand up
    6. Random wiggles and movements
  7. When the button text changes to "Start Recording" again, other buttons should become enabled
  8. OPTIONAL: If you want to save your recording to be used later, click the "Save Recording" button, this is currently mostly helpful for debugging purposes, to load recordings later, they must be placed in a subdirectory titled "LoadRecordings" within the SlimeVR server directory
  9. To calculate your body proportions from the recording (current or saved) press the "Auto Adjust" button; you should be able to see new values for the lengths of your body reported in meters
  10. To use the calculated values, press the "Apply Values" button. If the values do not look right, you can try recording again

How does it work?

Skeleton auto-configuration works by recording movement data and simulating that movement rapidly while gradually adjusting the bone lengths. When adjusting bone lengths, the algorithm measures the amount the feet slide to know whether it's doing better or worse with each adjustment. By iterating over the data multiple times, the algorithm is able to obtain reasonable bone length values with minimal foot sliding.

The skeleton auto-configuration algorithm uses classic machine learning technique called gradient descent to acquire the bone length values. First, many samples of movement data are recorded, then using gradient descent, the algorithm gradually adjusts the bone lengths to minimize the error of foot sliding. Error is calculated through multiple different methods, but generally it is formulated to retain the headset's reported height, "average" human body proportionality, and reduce the amount that the feet slide during movement.

Almost all of the algorithm's internal values are exposed through the config file, read the following Configuration Documentation section to learn more.

Configuration Documentation

All configuration options should be placed in the vrconfig.yml file and are sub-configs to autobone, for example:

zoom: 1.0
virtualtrackers: 3
autobone:
    sampleCount: 1000
    slideErrorFactor: 1.0
    calculateInitialError: True
Config Option Value Type Default Value Description
sampleCount Integer 1000 The number of pose samples to record
sampleRateMs Long Integer 20 The millisecond interval between each sample to record
saveRecordings Boolean False When true, recordings will automatically be saved
minimumDataDistance Integer 2 The minimum distance between samples to be used while adjusting
maximumDataDistance Integer 32 The maximum distance between samples to be used while adjusting
epochCount Integer 5 The number of epochs (full loops) to iterate over the data
adjustRate Float 2.5 The factor to adjust the values by each iteration
adjustRateDecay Float 1.01 The factor to decay the adjust rate by each epoch (divided by this value)
slideErrorFactor Float 1.0 The factor of which the foot slide error is used in the error calculation
offsetErrorFactor Float 0.0 The factor of which the foot offset error is used in the error calculation
proportionErrorFactor Float 0.2 The factor of which the body proportion error is used in the error calculation
heightErrorFactor Float 0.1 The factor of which the body height error is used in the error calculation
positionErrorFactor Float 0.0 The factor of which the absolute position error is used in the error calculation
positionOffsetErrorFactor Float 0.0 The factor of which the absolute position offset error is used in the error calculation
calculateInitialError Boolean True When true, the initial error over the data is reported as epoch 0
manualTargetHeight Float -1.0 The height to use for the height error calculation, this is calculated automatically when negative but can be overrided with this when set to a positive value in meters

created by Butterscotch!#7878