ITEM Lab — Kihyun Louis Park

Overview

Making the untouchable system easier to learn.

Buttons in VR lack physical contact, creating uncertainty around the exact moment and position of interaction.
We designed an adaptive feedback system that accelerates user adaptation, reducing the learning curve.

Cut drift by ~86% With adaptive feedback, users' press position stayed nearly flat and consistent from the previous trials.

Locked in early Adaptive haptic feedback model converged fast and held while others(e.g. visual only) were still adjusting.

Problem

Stuck in a loop of trial and error without any progress.

In virtual reality, users easily fall in a loop of overshoot* and underreach* with no idea of where to press exactly.
Over time, this erodes their sense of control and adds fatigue based on studies on button-press neuromechanics.

Overshoot : pushing past the contact point.

Underreach : stopping before the contact point.

Why Adaptive

More to feel, less to figure out.

Instead of an alert sound or fixed vibration, the feedback gives spatial cues that guide the hand to the contact point based on each user's behavior. As a result, no one has to spend time figuring out where to press exactly.

Fixed Haptic Feedback? In that case, the system has to choose the feedback spot. but... how?

Solution

Adaptive Haptic Feedback

The feedback point learns from each user's presses and shifts to match, instead of staying fixed. Ultimately, the user can learn the contact point faster and with less effort.

Discussed Topic (but not implemented) Inspired by a physical keyboard button, we thought about making a different level of vibration intensity based on which contact point the user is at. But we agreed that it would be too much for this project and left it for future work.

Experiment Design

How did we measure "learning"?

We split it into two different questions that can be answered with numbers.

μ as Where Did They Learn? If the average press position stops drifting across trials, the user has settled on a spot. Learning is complete.

σ as Did They Get Confident? If the spread between tries keeps shrinking, the user is pressing more consistently. Confidence is still rising.

5 Different Feedback Conditions (6 Participants)

Visual only baseline · the current VR norm
Vibrotactile Fixed Adaptive
Visual + Vibrotactile Fixed Adaptive

Result

Cutting drift & locked in early

Adaptive feedback settled the press position and held its consistency from the very first trials.

7× less drift than visual-only

Adaptive −0.0008 settled, stayed flat

Visual only +0.0058 kept drifting, never settled

Adaptive model converged early and held

Adaptive +0.0002 already tight early, flat after

Visual only −0.0019 still narrowing at the last trial

Constraints & Trade-offs

What we had to give up during this project

Small Experiment Sample We could only conduct 6 user studies. But we treated every result as a signal toward a bigger picture. To get more data, we ran a simulation for the conference participants.

Fine Tuning Adapt too aggressively, and it creates noise. Adapt too slowly, and the user never feels it learning. We picked a number(16) that felt stable in testing, but we didn't yet find the optimal one, and it likely differs per person.

Validation

We presented and validated at IEEE WHC 2025

This adaptive feedback system was presented at an IEEE WHC 2025, validating the approach of using adaptive haptics to improve learning in VR interaction. The project itself received positive feedback and was selected as a finalist!

Reflection

What I learned from this project

At first, I assumed the hard part was designing the interaction, like how to make buttons feel better in VR. The real lesson was upstream: if you can't define what "better" means as a number, you'll end up arguing about just feelings. Splitting a vague goal like "learning" into μ and σ is what turned a subjective debate into a testable one and that's the habit I'm taking into product work.