Wearable Sensors Reveal the Reality Gap:
A Full-Body Motion Dataset for Validating VR Cycling Simulators

Street2Simulator — Openly available benchmark dataset
Jonas Pöhler · Antonia Vitt · Florian Wolling · Florian Michahelles · Kristof Van Laerhoven
10 participants 6 IMU sensors per session 120 Hz sampling rate 1.4 km urban route · Vienna 20 total sessions Open access

About This Dataset

Virtual reality cycling simulators are increasingly used in research on urban mobility, rehabilitation, and road safety, yet their biomechanical fidelity remains under-validated due to the lack of standardised, publicly available datasets. Street2Simulator addresses this gap by providing synchronised, full-body inertial measurement unit (IMU) data from 10 participants who each cycled an identical 1.4 km urban route both in a motion-enabled VR simulator and in the real world.

Six body-worn sensors captured head, torso, arm, and leg movements at 120 Hz, enabling detailed comparisons of pedaling rhythm, limb coordination, balance control, and head orientation across environments. Egocentric video was recorded alongside the IMU data in both conditions.

Key finding: VR successfully replicates fundamental locomotor patterns such as pedaling cadence and bilateral leg coordination. However, significant differences emerge in upper-body dynamics — participants showed markedly greater torso rotational variability in VR compared to real-world cycling (26.8 vs. 12.4 deg/s; p = 0.020, Cohen's d = 0.96), indicating altered balance strategies in the simulator. IMU signal distributions shift from Cauchy-family (heavy-tailed) in VR to logistic/normal in the real world, revealing a qualitative change in motor control that mean-level comparisons alone cannot capture.

Study Design

VR Condition

  • Stationary bicycle on ±5° lateral tilt platform
  • Oculus Quest 2 HMD (90 Hz, ~97° FoV)
  • Unity3D virtual replica of the real-world route
  • Speed feedback via pedaling cadence (smart trainer)
  • Mean session duration: 153 s (SD = 27)

Real-World Condition

  • Same bicycle model on a paved urban road in Vienna
  • 1.4 km route: bike paths, intersections, tram tracks
  • Self-selected comfortable pace
  • Pupil Labs Core egocentric camera
  • Mean session duration: 354 s (SD = 54)

Participants

  • N = 10 complete trials (11 recruited, 1 excluded)
  • 9 male, 2 female; mean age 24.4 years (SD = 3.7)
  • Cycling experience: mean 2.09/5 (moderate)
  • All sessions in one week (October 2024)
  • Ethics approved: University of Siegen LS_ER_03_2023

Data Streams

  • Raw tri-axial accelerometer (±8 g)
  • Raw tri-axial gyroscope (±2000 deg/s)
  • Egocentric video at 30 fps
  • Simulator Sickness Questionnaire (SSQ)
  • Presence, achievement, and affect ratings

Sensor Positions

Six wireless IMU nodes (InvenSense MPU-6050, 120 Hz) were worn simultaneously using elastic straps. Synchronisation across nodes was achieved via a shared radio timestamp; clock drift did not exceed 10 ms per session.

PHR
Head
Right temporal region. Captures head orientation, yaw range, and visual search behaviour.
RKL
Torso
Sternum / upper back. Primary outcome site for postural sway and balance control.
RHL
Left Forearm
Captures handlebar-transmitted vibration and active steering movement.
RBL
Right Forearm
Mirror of left forearm; enables bilateral upper-limb comparison.
PVL
Left Shank
Lower leg. Used for pedaling cadence extraction (gyroscope peaks = downstroke).
RBR
Right Shank
Mirror of left shank; bilateral coordination quantified via cross-correlation.

Key Results

Torso Angular Rate SD — Primary Outcome
VR: 26.8  vs.  Real: 12.4 deg/s
paired t(8) = 2.889, p = 0.020*, d = 0.963 (large)
Head Yaw Range
VR: 338.9°  vs.  Real: 220.8°
paired t(7) = 2.726, p = 0.026*, d = 0.909 (large)
Overall Impression (0–4 scale)
VR: 2.0  vs.  Real: 3.4
paired t(10) = −4.404, p = 0.001*, d = −1.328
Composite Achievement Score
VR: 5.0  vs.  Real: 7.3
paired t(10) = −4.077, p = 0.002*, d = −1.229
Pedaling Frequency
VR: 0.59  vs.  Real: 0.82 Hz
paired t(8) = −1.606, p = 0.147 (n.s.), d = −0.535
Simulator Sickness (SSQ Total)
M = 6.2 — Mild
Below clinical threshold; no correlation with head motion

* p < 0.05  |  Red = significant difference VR vs. Real  |  Orange = trend / underpowered  |  Green = null / benign result

Dataset Browser

Each row below corresponds to one participant. Both the real-world ride and the VR simulator session are available as interactive time-series viewers with synchronized accelerometer and gyroscope traces for all six body positions.

Participant Age Gender Cycling Exp. View Data
P01 21 m 3/3 Real Ride Simulation
P02 25 m 2/3 Real Ride Simulation
P03 27 m 3/3 Real Ride Simulation
P04 32 m 2/3 Real Ride Simulation
P05 22 m 1/3 Real Ride Simulation
P06 22 m 3/3 Real Ride Simulation
P07 22 m 2/3 Real Ride Simulation
P08 30 f 2/3 Real Ride Simulation
P09 23 f 1/3 Real Ride Simulation
P10 22 m 2/3 Real Ride Simulation

Each viewer shows tri-axial accelerometer and gyroscope signals for all six body positions. Traces are synchronised — hovering over one plot moves the cursor across all plots simultaneously. Participant metadata (age, gender, cycling experience) is partially anonymised.