PredicTouch: A System to Reduce Touchscreen Latency using Neural Networks and Inertial ...
summary
PredicTouch uses wrist, finger, or stylus IMUs plus a multi-layer feedforward neural network (500/400/300 units, AdaGrad, outputs X/Y every 16 ms) to predict future touch positions and mask display lag. The team collects 1M touch samples matched to 28M IMU samples, trains models per-IMU combination, and finds wrist IMUs best for finger and stylus IMUs best for pens. Prediction error drops (example averages ~9, 21, 35 pixels across horizons) and throughput rises ~15% (finger) and ~17% (stylus).
AI-written summary · about
chapters (10)
- 0:00Introduction to touch latency
- 1:22Previous latency research
- 2:02Existing reduction methods
- 4:35The PredictTouch system
- 5:52Data collection study
- 7:29Neural network architecture
- 9:37Prototype demonstration
- 10:45Evaluation of results
- 12:57Conclusion and future work
- 13:49Q&A session
related talks
18mInfiniTouch: Finger-Aware Interaction on Fully Touch Sensitive Smartphones
19mATK: Enabling Ten-Finger Freehand Typing in Air Based on 3D Hand Tracking Data
16mInvestigating the feasibility of finger identification on capacitive touchscreens using deep ...
08mSo Predictable! Continuous 3D Hand Trajectory Prediction in Virtual Reality
08mDesigning a Personalised Sensor Glove Using Deep-Learning
19mGroup Touch: Distinguishing Tabletop Users in Group Settings via Statistical Modeling of Touch Pairs
11mLitePred: Transferable and Scalable Latency Prediction for Hardware-Aware Neural...
14mLeveraging Dual-Observable Input for Fine-Grained Thumb Interaction Using Forearm EMG