• Home
  • Dataset Description
  • ‣ Download
  • ‣ Experimental Setup
  • ‣ Experimental Protocol
  • ‣ Data Description
  • ‣ Known Issues
  • License
  • ‣ License Terms
  • ‣ Participant Consent
Puzzle-STAMPS Puzzle-STAMPS
  • Home
  • Dataset Description
    Download Experimental Setup Experimental Protocol Data Description Known Issues
  • License
    License Terms Participant Consent

Dataset Description

Puzzle-STAMPS — A Multimodal Motion–Physiology–Speech Dataset

Download Experimental Setup Experimental Protocol Data Description Known Issues

Experimental Setup

The experimental environment was designed to create a collaborative escape-room experience within a controlled laboratory setting. Teams were given no predefined roles and were free to determine their own collaborative dynamics throughout the experiment. The setup comprises a primary puzzle box, a secondary toolbox, and a hint-delivery system, all integrated with a unified recording pipeline.

The study comprised thirteen unique puzzles organized into eight timed segments. The third segment featured five independent sub-puzzles whose solutions unlocked a final integrating puzzle, while the other seven segments contained a single puzzle each. The tasks demanded diverse coordination skills: shared visual search, spatial reasoning (e.g., path-finding and jigsaw puzzle assembly), optional parallel workflows in segment 3, and real-time teamwork, such as navigating an electronic maze where the display and controls sit on opposite sides of the puzzle box. Some puzzles also allowed brute-force solutions. This variety was designed to elicit diverse leadership and coordination behaviors.

Puzzle Box

The primary task centered on a modified Deadly Hallows electronic puzzle box (Indestroom, HU). The box presents interactive puzzles across its four sides and top surface, combining digital displays (screens, LED indicators) with physical mechanisms (buttons, dials, rotating disks, locks). The box’s Unity-based control application was extended to interface with the recording pipeline, enabling automatic logging of puzzle state transitions (start, solved, failed).

Puzzle box – Side 1
Side 1 © Indestroom
Puzzle box – Side 2
Side 2 © Indestroom
Puzzle box – Side 3
Side 3 © Indestroom
Puzzle box – Side 4
Side 4 © Indestroom
Puzzle box – Top
Top© Indestroom

Toolbox

To expand the problem-solving space, the setup was augmented with a secondary toolbox containing seven numbered compartments, each secured by one to three four-digit combination locks. As teams solved specific puzzles on the main box, motorized trapdoors opened automatically to reveal strings of letters, numbers, and emojis taped to their interior. Participants had to communicate or memorize these codes to unlock the corresponding compartments in the toolbox, which contained essential physical artifacts needed for subsequent puzzle stages.

Toolbox with seven locked compartments
Toolbox

Hint & Timer System

A monitor near the toolbox displayed a synchronized countdown for the current segment and provided time-triggered written hints, delivered at three escalating levels from directional nudges to near-explicit guidance. An audible signal played from a dedicated speaker whenever new information or hints appeared on the screen. If a segment timer reached zero, the system automatically skipped any unsolved puzzles to ensure that the session did not exceed the 180-minute limit. The puzzle box’s Unity-based control application was extended to interface with our recording pipeline, enabling automatic logging of state transitions and skip events with timestamps.

Laboratory Layout

The experiment took place in a laboratory at the Max Planck Institute for Intelligent Systems, Stuttgart. The room was arranged to host one team of four participants at a time to gather around the puzzle box, which was positioned centrally.

Annotated floor plan showing sensor positions, camera angles, and puzzle box layout
Laboratory floor plan

Room Configuration

The room was configured as follows:

  • Puzzle box: Centrally located; participants stood or sat around it during the session.
  • OptiTrack Prime 17W cameras: Eight cameras mounted at ceiling height, providing full-room motion capture at 360 Hz via Motive software.
  • Room video cameras: Five fixed Reolink RLC-511 CCTV cameras capturing complementary exocentric viewpoints at 25 fps.
  • Experimenter station: Located outside the main interaction area, hidden behind a curtain.

Sensor Placement on Participants

Each participant was simultaneously instrumented with the following sensor systems prior to the session.

Movement & Localization

Sensor Specification Role
Pozyx Creator UWB tag (6-anchor RTLS) Planar position + 9-DOF IMU, variable refresh rate Coarse real-time room-level localization. Each participant wore an ultra-wideband (UWB) tag tracked by a six-anchor Pozyx Creator real-time location system, recording planar position and 9-DOF inertial data.
OptiTrack Prime 17W optical system (8 cameras) 360 Hz, Motive software High-precision head pose and orientation. Each participant wore a motion cap fitted with the UWB tag and reflective markers; the cap was calibrated via marker-equipped glasses to align with the participant’s visual midpoint, providing an accurate approximation of egocentric gaze and orientation during puzzle interaction and social exchange.

Physiological Garments

Each participant wore a Healer R2 smart garment (L.I.F.E. Italia S.r.l.), a clinical-grade wearable providing cardio-respiratory signals comparable to standard medical equipment. The sleeveless close-fitting garments are machine-washable and come in a wide range of sizes for men and women to ensure stable sensor-to-skin contact without hindering physical movement. Each garment has its own on-body logger, which synchronized its internal clock to the local network. These physiological channels enable the study of interpersonal synchrony, stress reactivity, and arousal dynamics in collaboration and leadership scenarios.

Signal Sample rate Details
ECG 500 Hz Three-lead electrocardiogram.
Respiration 50 Hz Three strain sensors: thorax, xiphoid, and abdomen.
SpO2 1 Hz Blood oxygen saturation.
Skin temperature 1 Hz Surface skin temperature.
9-axis IMU 100 Hz Upper-back inertial signals (accelerometer, gyroscope, magnetometer) integrated in the garment.

Audio

Component Details
Rode Lavalier II microphone Close-talk lapel microphone attached to a surgical mask worn by the participant, oriented toward the wearer’s mouth to maximize signal-to-noise ratio.
Rode Wireless Go II digital transmitter Wireless transmission of individual audio streams to the main acquisition computer. Each participant had a separate channel, yielding one close-talk recording per participant.

Video

Component Details
5× Reolink RLC-511 CCTV cameras Fixed cameras capturing complementary exocentric viewpoints of the interaction area. Streams were captured continuously on a dedicated Synology DiskStation DS1520+ NAS.

Synchronization & Recording Pipeline

All sensing systems were isolated from the internet and synchronized to the main acquisition computer, which served as the local NTP server. The main computer captured both the OptiTrack streams and the transmitted lapel microphone audio, while a dedicated Raspberry Pi 5 recorded the puzzle-box game-state logs and the Pozyx RTLS data. Each physiological garment has its own on-body logger, which synchronized its internal clock to the local network. The CCTV streams were captured continuously on a dedicated Synology DiskStation DS1520+ NAS.


Next: Experimental Protocol →

© 2026 Max-Planck-Gesellschaft - Imprint - Privacy Policy - License
© 2026 Max-Planck-Gesellschaft