VirteAnalysis

This is the front page for the Virte Experiment documentation and analysis.

Links

• VR06Paper—Preparation for a paper on Virte4 that was not submitted.
• VirteDataset—Documentation relating to the the Matlab dataset and code used in our analysis.
• PerPathMeasures—Measurements describing a single path. These are written to a CSV file so that we can do a statistical analysis.
• VirteTodo—My own to-do list of Virte-related tasks.
• Virte1BadTrials, Virte3BadTrials—List of anomalous trials that should be thrown out.
• StartingPoint, StoppingPoint, TargetPlane—Has some helpful diagrams of how we define the beginning and end of an approach.
• Virte4TopDownView—Bird’s eye view of subject 4001 in the 5 conditions.

Topics and Questions relating to the Virte Experiment Analysis

Some of these are technical aspects, others are behavioral.

• Space-curve Representation
  • PathsIn3Space
  • FrenetFrame
  • DerivativeSignal
  • Using forward (tangential) velocity and forward (tangential) acceleration.

• PathAlignment
  • Fixed StartingPoint
  • Variable StoppingPoint

• Motion Paths as 1-d Signals
  • Projected onto axis orthogonal to TargetPlane.
  • Parameterize vs time or distance? Both seem to be useful.
  • We have different path lengths.
    • So scale them or truncate them to match?

• PerPathMeasures
  • Identifying CriticalPoints
  • PCA
  • FactorAnalysis

• Attempts at Multidimensional Scaling.

Joystick as a Means of Locomotion

• Stopping point is hard to identify. Subjects often come to complete stop and then nudge the controller to correct. We want to measure the final distance-to-wall in order to gauge how close people were trying to get. So we care about distance after all the nudging is over. Yet we want to measure subjects’ ability to brake and stop where they wanted to stop. So we want to measure the initial stopping point, before the nudging and other corrections.

• What are the common usage patterns in Virte1,3?
  • View unfiltered joystick velocity data.
  • Undershoot and then nudge-up to final stopping point.
  • Overshooting. Can we identify these?
  • Full-throttle followed by full-stop, versus more nuanced control.

• Identify a small handful of subjects who were expert joystick users (based on their smooth usage of the joystick?).
  • Characterize how we’re able to recognize the expert behavior.
    • Smooth usage of the joystick.
    • Consistency.
  • Would be great if we could show that, by the end of the trial, even non-expert users began to look more like the behavior we see in the expert users.

• How a subject’s behavior changes during the course of the trial. Mastering the controls.
  • Visualize by overlaying all targets/corners for a single condition, single subject.
  • Hope to find that behavior becomes smoother and more consistent.

• What are the common usage patterns in Virte4?
  • Straight-line
  • L-shape (sharp corner)
  • Sliding off wall
  • Smooth curve
    

What role does Tau play in braking behavior?

• Based on perceived distance-to-target.
• What is Tau-dot?

Performance metrics.

• In what aspects of the task can we say that subjects performed well or poorly?
  • Final distance to taret.
    • Confounded by perceptual differences.
    • Confounded by differences in interpretation of the task.
  • Overall Smoothness of motion. meausre by aggregate jerk
  • Could we measure how well a person has mastered the locomotion controls?
    • Would help in user studies.

Where did people stop, relative to the target/corner?

• Visualize this by scatter-plots from an over-head perspective.
• Is there a significant distancde between

Statistical Questions

• Are we justified in considering the 6 corners/targets as multiple trials of the same condition, or not?
• Does passive haptics make a difference?
  • differences among corners
  • differences among targets
  • consistency of participants over corners/targets
• Are corners different from targets?
  • differences between of corners and targets (easier/harder depending on whether we can consider the 6 as multiple measures of the same thing)

Overall differences in speed

• VR was slower than Real/Cowl.
• No difference between Real and Cowl.
• Joystick speed probably depends on how we set the parameters.
• WIP was slower, but speed probably depends on how we set the parameters.
  • Some people had nearly constant speed (when low-passed), which is a result of regularly-placed steps with no missed steps.

Other

• How can we identify the ways in which all paths differ?
• How can we identify whe ways in which all paths are similar?
  • How are the per-path measures distributed?
  • “Eye-ball it.”
    • What are the observed characteristics of derivative curves
    • What are the observed characteristics when replaying the log.

• Pointwise comparison. comparing path shape.

VirteTerminology

Wall

A wall-mounted visual target in the virte3 experiment. Walls number 1—10.

Corner

A corner in the virte3 experiment. Corners number 11—18.

Target

Generic name for a wall or corner. Virte3 has 18 walls.

Target Plane

This term applies to both walls and corners. For walls this is a physical plane, for corners it is imaginary.

Approach

This refers to the portion of the data leading up to a single wall or corner.

Trajectory

Spacecurve data of the tracked point’s motion through space. (We used to call it a “path”, but “trajectory” seems more self-explanatory.)

Velocity Profile

The subject’s velocity, plotted versus either time or position. This provides the best 1-d visualization of the subject’s locomotion behavior.

Tau Profile

The subject’s Tau, plotted versus either time or position.