Virtual Reality-Simulated Interaction Between Micro-Mobility Vehicles and Pedestrians: A Biomechanical Analysis of Human Gait and Movement Responses

Technical Summary: Virtual Reality-Simulated Interaction Between Micro-Mobility Vehicles and Pedestrians

Problem Statement
The rapid adoption of electric scooters (e-scooters) has intensified interactions between pedestrians and micro-mobility vehicles on shared sidewalks and crossings, elevating collision risks. While existing literature relies heavily on trajectory-based observations and field data, there is a significant gap in understanding the detailed biomechanical gait responses of pedestrians during these interactions. Current studies lack insight into how perceived collision risks trigger specific, reflexive motor adaptations in human gait, particularly in controlled environments that isolate these variables from real-world confounding factors.

Methodology
This study employed an immersive Virtual Reality (VR) environment combined with markerless pose estimation to quantify pedestrian biomechanical responses.

• Participants: The final analysis included 12 healthy male university students (aged 21–23) screened for musculoskeletal disorders and VR tolerance.
• Experimental Design: Participants walked through four distinct VR scenarios presented via a Meta Quest 3 headset:
  1. Baseline: Normal pedestrian-only walking.
  2. Encounter: E-scooter passing at speeds of 10–25 km/h.
  3. Crossing: E-scooter crossing the pedestrian's path.
  4. Near-Crash: A simulated "crash-in" scenario representing an imminent collision threat.
• Data Acquisition: Sagittal-plane video was recorded at 30 fps (1080p) using a GoPro HERO 12 mounted at hip height.
• Analysis Pipeline:
  • Spatiotemporal Parameters: Extracted manually using Kinovea software, focusing on two gait cycles (GC) per trial. Metrics included step length, gait cycle time, velocity, stance phase, and swing phase.
  • Joint Kinematics: Markerless pose estimation was performed using the OpenPose 25-point model (BODY_25) to detect joint coordinates. A custom JSON-based pipeline extracted left knee joint trajectories. Knee angles were calculated based on hip-knee-ankle coordinates and smoothed using a Savitzky–Golay filter to reduce noise.
• Statistical Analysis: Data were analyzed using ANOVA, MANOVA, and non-parametric tests (Student's t-test) via GraphPad Prism to compare scenarios.

Key Results
The study identified statistically significant biomechanical adaptations in response to increasing levels of perceived threat:

• Spatiotemporal Changes:
  • Step Length: Significantly reduced in high-stress scenarios. Step length decreased from a baseline of 226.5 cm to 204.7 cm in near-crash simulations ($p < 0.001$).
  • Gait Velocity: Velocity showed a significant reduction in the near-crash scenario (71.6 cm/s) compared to the baseline (84.5 cm/s) and the initial encounter scenario (92.6 cm/s), with a highly significant difference between the encounter and near-crash scenarios ($p = 0.001$).
  • Timing: While overall gait velocity and step time changes were not always statistically significant across all comparisons, participants consistently exhibited shorter stance phases and longer swing phases during stressful encounters.
• Kinematic Adaptations:
  • Knee joint angle trajectories revealed altered flexion-extension patterns. Visual analysis indicated "jitteriness" in knee angle progression during crossing and near-crash scenarios (Scenarios 3 and 4), suggesting restricted knee motion and reflexive postural adjustments.
  • The study noted that the right knee joint data was less reliable due to occlusion by the contralateral limb in the sagittal view, limiting the analysis primarily to the left lower limb.

Key Contributions

1. Biomechanical Quantification of Risk Perception: The study demonstrates that pedestrian gait adapts measurably to micro-mobility threats even in the absence of physical contact. These adaptations (reduced velocity, shortened step length, altered stance/swing ratios) serve as objective indicators of perceived collision risk.
2. Methodological Validation: It validates a workflow combining consumer-grade VR headsets, 2D action cameras, and open-source pose estimation (OpenPose) for quantifying human gait in simulated traffic environments. This offers a scalable, non-invasive alternative to traditional marker-based motion capture systems.
3. Threat-Dependent Modulation: The findings confirm that gait modulation is graded; the most pronounced changes occur during near-crash scenarios, supporting the hypothesis that pedestrians employ cautious, stabilizing strategies when collision risk is imminent.

Significance and Claims
The authors claim that these findings provide a foundational understanding of how pedestrians react to emerging micro-mobility modes. The identified gait adaptations can serve as sensitive biomarkers for collision risk, potentially supporting the development of:

• Proactive Safety Measures: Infrastructure designs and operational policies that mitigate conflicts between e-scooters and pedestrians.
• Intelligent Control Systems: The integration of these biomechanical indicators into intelligent micro-mobility control systems to anticipate pedestrian behavior.
• Intervention Testing: A scalable template for evaluating safety interventions (e.g., speed limits, warning systems) using VR-based gait analysis.

The paper maintains a modest stance, acknowledging limitations such as the small, homogeneous sample size (healthy young males), the restriction to two gait cycles, and the reliance on 2D sagittal-plane analysis which may miss multi-joint coordination or ankle kinematics. The authors emphasize that while the VR environment mimics realism, it may induce motion sickness, and future work should consider diverse populations and 3D motion capture for comprehensive analysis.