Lost in Corridors: Modeling and Mitigating Spatial Disorientation by Sensing Environmental Characteristics and User Behavior

Paper Title

Lost in Corridors: Modeling and Mitigating Spatial Disorientation by Sensing Environmental Characteristics and User Behavior

Publication Info

• Topic area: Spatial cognition and adaptive navigation systems.
• Keywords: Spatial disorientation, indoor navigation, geometric symmetry, feature similarity, multimodal behavior, CNN-BiLSTM, VR experiments, adaptive systems, cognitive state detection, navigation aids.

Background and Problem

• Problem / challenge: Current navigation systems are reactive and generic, failing to address the root environmental causes of spatial disorientation or integrate the user’s cognitive state.
• Significance: Spatial disorientation in repetitive indoor layouts leads to inefficiency, frustration, and inhibits spatial learning. Addressing this issue is critical for improving navigation systems in complex environments like hospitals and malls.
• Motivation and related work: Prior research has explored spatial cognition, cue integration, and context-adaptive systems but lacks robust models for detecting the "getting lost" state or understanding how geometric symmetry and feature similarity interact to cause disorientation.

Solution

• Proposed approach: A proactive, context-aware navigation framework integrating dual context-awareness (environment and user) and content-adaptive aid strategies.
• Novelty:
  1. A quantitative model linking environmental characteristics (geometric symmetry and feature similarity) to user navigation behavior and cognitive load.
  2. A CNN-BiLSTM model for detecting the "getting lost" state using multimodal behavioral data with >90% accuracy.
  3. Design principles for adaptive navigation systems that dynamically balance efficiency and spatial learning.
• Procedure and key techniques:
  • Conducted a VR experiment (N=40) manipulating geometric symmetry and feature similarity.
  • Collected multimodal behavioral data (e.g., body motion, eye tracking) and operationalized the "getting lost" state using heuristic behavioral proxies.
  • Developed and validated computational models (CNN-BiLSTM, LSTM, 1D-CNN) for real-time detection of disorientation.

Results

• Concrete findings:
  • High feature similarity increased hesitation duration by 370% in symmetrical environments.
  • Geometric symmetry primarily affected exploratory body rotation, while feature similarity impacted navigation outcomes.
  • Dual-cue failure triggered a strategic shift from active reorientation to locomotor compensation (e.g., wall-following).
• Advantage over baselines:
  • CNN-BiLSTM model achieved >90% accuracy and an F1-score of 0.76 for detecting the "Lost" state, outperforming prior unimodal models (70–75% accuracy) and simpler classifiers (F1-score: 0.23).
• Experiments / evaluation:
  • 2×2 factorial design manipulating geometric symmetry and feature similarity in VR environments.
  • Metrics included path tortuosity, hesitation duration, body rotation variability, and cognitive load.
  • Models evaluated using accuracy, macro-F1, and class-specific F1 scores.
• Limitations and future work:
  • Proxy-based labeling lacks direct cognitive validation.
  • Findings may not generalize to older or less spatially skilled populations.
  • Single-room VR paradigm limits applicability to large-scale navigation tasks.
  • Real-world deployment requires adaptation to AR systems with limited body motion sensors.

Summary

This study addresses spatial disorientation in repetitive indoor layouts by modeling the interaction of geometric symmetry and feature similarity with user behavior and cognitive load. A CNN-BiLSTM model was developed to detect the "getting lost" state with high accuracy using multimodal behavioral data. Results reveal a functional separation between geometric and featural cues and a dynamic cost-benefit trade-off driving navigation strategies. The proposed framework offers design principles for adaptive navigation systems that integrate dual context-awareness and content-adaptive aids, aiming to balance efficiency and spatial learning. Future work should extend these findings to diverse populations and real-world applications.