Joystick accuracy test

I got a few people to run a program I made to test their accuracy with the thumbstick on a gamepad.

What kicked off this train of thought, was I was considering radial UI as an interface. How many options could you reasonably include in a radial menu? That would depend on how accurate you could count on people being.

The test showed the user an arrow indicating a direction they needed to match. They then needed to point their thumbstick in the same direction as the arrow and press a button.

An example of what the user saw during testing

I put each used through three phases. The first and last phase showed no visual feedback about the direction their thumbstick is facing. The second phase did show the direction they are pressing as another arrow and they needed to match the two arrows. Repeating with no visual feedback in the final phase was to test whether accuracy would improve after the second phase.

To ensure everyone was tested on a full range of directions, I generated 30 directions for each phase, each within a different 12 degree range, and then shuffled them. So every 12 degree segment is tested once per phase.

You can see here, phase 2 was much more accurate when they got visual feedback.

The mean average for degrees inaccuracy in phase 1 was 10.9 and in phase 2 was 10.4. Median went from 9.2 to 8.8. So it doesn’t feel like the training phase lead to a huge improvement.

I got curious about whether there were certain directions more prone to clockwise or counter-clockwise error, but my first shot at visualizing that was difficult to understand.

This was cleaner. Instead of connecting the target and goal angles, I created lines at the target angle, and extended it outward for clockwise error, and inward for counter-clockwise error.

Maybe there’s a small cluster of rightward at the top and bottom right. There may or may not be a slight rightward skew, but it’s a small enough sample it might be a coincidence.

This is the average input direction for each user, compared to the average target direction. If this isn’t a coincidence, I wonder if the slight right offset is from using the left thumbstick, and tests with the right thumbstick would show the opposite?

Here we have each subject (named A to H) and a box plot of their inaccuracy for first and third phase.

In terms of improving after the test phase, I found as many subjects were worse during the third phase, just as many as improved. For accuracy, if we had a 5-section radial menu, it would require accuracy within 36 degrees, which would fit most of this. 6-section radial would require 30 degree accuracy, which would start excluding a few people here. In practice, I’m sure I’d add visual feedback for accuracy, but it’s an interesting experiment.

Maybe also I’ll try a future test where, instead of arbitrary direction on the circle, visually divide it into segments like a radial menu. Then see how people fare with 5 segments, then 6, then 7 and so on.

Recent Posts