The Effect of Latency on Movement Time in Path-steering

Title of the Paper

The Effect of Latency on Movement Time in Path-steering

Paper Information

• Domain: Human-Computer Interaction (HCI), Modeling of Manipulation Performance
• Keywords: Human motor performance, movement time prediction, graphical user interface, path steering, latency effects, experimental evaluation, Fitts' law, steering law

Research Background and Problem

• Problems and Challenges:

  • In current graphical user interfaces, the "inevitable latency" between the user's cursor operations and the interface response may reduce execution efficiency.
  • While extensive research has analyzed the effects of latency on target pointing tasks and developed mathematical models, the correlation between latency and movement time (MT) in path-steering tasks lacks systematic study.
  • There is no universal theoretical model to predict the impact of latency on path-steering tasks, requiring separate experiments for different latency conditions in existing studies.

• Research Significance:

  • Path-steering is a critical interaction method in GUIs, such as navigating cascading menus or performing drawing tasks.
  • Developing mathematical models for high-latency scenarios can not only deepen the understanding of user behavior but also enable more precise interface design to improve user experience.

• Research Motivation: To propose a mathematical model that explains and predicts the impact of latency on MT in path-steering tasks, addressing the gap in latency effect modeling in existing studies.

Solution

• Methods and Solutions:

  • A revised predictive model of the traditional steering law is proposed to quantify the effect of latency on MT in path-steering tasks.
  • The model is theoretically derived and validated through five experimental tasks: goal crossing, linear paths, circular paths, paths with varying widths, and target-pointing tasks.

• Innovations:

  • For the first time, latency is introduced as a linear constraint factor based on Fitts' law and the steering law, and a new mathematical model is derived to predict MT in path-steering tasks.
  • Cross-validation demonstrates that the revised model significantly outperforms traditional models in predicting MT.

• Key Implementation Steps:

  • Model Derivation: Based on Fitts' law and the steering law, latency (L) is linearly incorporated into the index of difficulty (ID). The formula includes total end-to-end latency (L_total).
  • Experimental Design: Five user tasks (goal crossing, linear paths, circular paths, paths with varying widths, and target-steering tasks) are designed to evaluate the model's predictive performance under various task types and latency conditions.
  • Data Analysis and Validation: The model's predictive performance and accuracy are assessed using AIC criteria, adjusted goodness-of-fit (R²), and cross-validation (LOOCV).

Research Findings

• Specific Results:

  • The revised model significantly improves MT prediction accuracy across all five experimental tasks, with adjusted R² values exceeding 0.94.
  • The model is effectively applicable within the realistic latency range (50-250ms), outperforming all baseline models.

• Advantages Over Existing Solutions:

  • Compared to traditional steering law models, the new model fully accounts for the impact of latency, resulting in more accurate predictions.
  • AIC metrics are significantly reduced, and cross-validation (RMSE) errors are also notably decreased.

• Experimental Results:

  • Task 1 (Goal Crossing): Incorporating latency significantly improves MT prediction, with the linear latency model (R² > 0.97) accurately predicting longer MT scenarios.
  • Task 2 (Linear Path Steering): The baseline model achieves R²=0.84, while the latency model improves to R²=0.98, with prediction errors significantly reduced (RMSE from 270ms to 90ms).
  • Task 3 (Circular Path Steering): In curved path steering, the latency model also demonstrates superior fitting performance, enhancing MT prediction consistency.
  • Task 4 (Variable-width Path Steering): The amplification effect of latency in narrow paths is clearly captured, with the revised model significantly outperforming the baseline in predictive performance.
  • Task 5 (Target-steering Combination Task): Comparisons of the best model's AIC and RMSE indicate that the latency gain term is particularly effective for complex combination tasks.

• Limitations and Future Directions:

  • Device limitations: The experiments primarily used mouse input and did not consider other input devices (e.g., VR headsets, gestures). Broader validation is needed in the future.
  • Latency range: Although the study covers the typical latency range of 50-250ms, extreme high/low latency scenarios require further validation.
  • Task complexity: The study does not fully address the effects of latency on multi-segment or highly intricate path-steering tasks.

Conclusion

This study proposes a novel quantitative predictive model for MT in path-steering tasks under latency conditions, significantly improving the effectiveness of existing steering laws across various interaction tasks. Future applications could extend to GUI design in VR/AR environments and multi-device collaborative scenarios, helping to mitigate the negative impact of latency on user experience.