Can a Computer Tell Differences between Vibrations?: Physiology-Based Computational Model for Perceptual Dissimilarity Prediction

Title of the Paper

Can a Computer Tell Differences between Vibrations? Physiology-Based Computational Model for Perceptual Dissimilarity Prediction

Paper Information

• Research Area: Human-Computer Interaction and Computational Modeling, focusing on physiology-based prediction of vibration perceptual dissimilarity
• Keywords: perceptual dissimilarity, computational model, bio-simulation modeling, haptic perception, vibration similarity, haptic interface, neural signals
• Publication Date: April 2023
• Conference/Journal: CHI 2023 (The ACM CHI Conference on Human Factors in Computing Systems)

Research Background and Problem

• Problems and Challenges:

  • Designing distinguishable vibration patterns requires costly subjective user ratings.
  • Existing methods for measuring vibration similarity or dissimilarity lack predictive accuracy, struggle to align with user data, or fail to generate structured perceptual spaces.
  • Current metrics have not been validated against user-perceived dissimilarity, limiting the flexibility of haptic interaction design.
  • There is a lack of high-performance tools for automated vibration design.

• Significance:

  • Accurately predicting user-perceived vibration dissimilarity can improve the efficiency and precision of haptic pattern design.
  • In virtual reality, wearable devices, and mobile devices, haptics is key to enhancing user immersion and improving information transmission.

• Motivation and Related Work:

  • Haptic design requires effective modeling of perceptual differences in vibration patterns to help designers reduce development time.
  • Many studies focus on mathematically-based signal processing models but lack reproducibility tied to physiological principles.
  • Related works like ST-SIM, RMSE, and SPQI can predict certain parameters but fail to provide comprehensive judgments on user-perceived similarity.

Solution

• Method/Model:

  • A physiology-based model (PM) is proposed, mimicking the human tactile system to predict perceptual dissimilarity in vibration pattern sets.
  • The model includes two parallel processing pathways: Neural Coding (NC) and One-dimensional Convolution (OC).
    • NC Pathway: Biologically models neural signal transmission, from mechanoreceptor activity in the skin to brain perception.
    • OC Pathway: Extracts rhythmic features using one-dimensional convolution processing.
  • By combining dissimilarity matrices generated by NC and OC, the model predicts perceptual dissimilarity in vibrations.
  • Eight parameters, including neural network layers and weight parameters, were optimized.

• Innovations:

  • Applied human tactile perception models (e.g., four-channel theory, neural signal transmission models) to vibration dissimilarity prediction.
  • Combined spectral and rhythmic characteristics to enhance prediction comprehensiveness and robustness.
  • Optimized model parameters using existing datasets and validated the model on multiple untrained and noise-influenced datasets.

• Implementation Steps and Key Techniques:

  1. Neural Coding (NC):
     • Used band-pass filters to simulate the vibration transmission process and mechanoreceptor activity.
     • Established a neural spike generation and propagation model, including excitation and inhibition processes.
  2. One-dimensional Convolution Pathway (OC):
     • Used moving average filters to extract rhythmic characteristics.
     • Calculated perceptual dissimilarity using Dynamic Time Warping (DTW).
  3. Model Training:
     • Optimized model parameters using six public datasets, fitted via Spearman correlation.
  4. Model Validation:
     • Validated on both trained and untrained datasets (including experimental data and measured acceleration data).

Research Outcomes

• Specific Results:

  • Validation on Training Datasets:
    • PM achieved an average correlation of 𝜌=0.79 in training datasets, outperforming six baseline metrics.
    • The generated perceptual space (via non-metric multidimensional scaling) exhibited the best structural similarity to user data (average alienation coefficient 𝐾=0.26).
  • Validation on Untrained Datasets:
    • PM achieved a correlation of 𝜌=0.67 on test datasets while generating perceptual spaces similar to user perception (𝐾=0.32).
    • Test datasets included designer-targeted vibration sets (IPS) and noise-influenced measured vibration sets (MPS), demonstrating PM's robustness to noise.
  • User Experiments:
    • User experiments on 12 different vibration patterns validated that PM outperformed other metrics in distinguishing perceptual rhythm and amplitude.

• Comparison with Existing Solutions and Advantages:

  • PM's predictive ability surpassed existing baseline metrics (e.g., RMSE, DTW, SPQI) across various datasets and parameter variations.
  • PM performed better in handling noisy data.
  • By integrating rhythmic and spectral characteristics, the model demonstrated higher flexibility in structured perceptual spaces.

• Limitations and Future Directions:

  • Limitations:
    • The model is limited to fixed-duration vibration inputs and lacks adaptability to varying input lengths.
    • The number of datasets used was limited, and the data range did not fully cover the diversity of vibration parameters.
    • The model assumes a fixed skin contact scenario (0.5mm probe) and does not account for other scenarios.
  • Future Improvements:
    • Expand the diversity of training datasets, including more untrained parameters.
    • Enhance the model's adaptability to different behaviors and contact modes, such as considering finger contact methods and device types.
    • Introduce advanced vibration propagation simulations to further enrich flexibility in user interaction.

Conclusion

This study proposes an effective method for predicting perceptual dissimilarity in vibration patterns by simulating the physiological model of human tactile perception. Experimental results demonstrate that PM not only accurately predicts user-perceived dissimilarity but also generates structured representations that align with user perceptual spaces. Therefore, this model has the potential to become a low-cost and flexible auxiliary tool in the field of vibration design, advancing the development of haptic interfaces.