BODYinTRANSIT-Lab: A Multimodal Dataset for Investigating Sensory-Driven Transformations in Body Perception and Affective Experiences
Authors: Mohammad Mahdi Dehshibi, Tomás Martínez Cortés, Amar D'Adamo, Daniel de la Prida, Luis-Antonio Azpicueta-Ruíz, Laia Turmo Vidal, Marte Roel Lesur, Joaquín R. Díaz-Durán, Alexander Väljamäe, Angel Sánchez, Temitayo Olugbade, Nadia Bianchi-Berthouze, Fernando Díaz-de-María, Ana Tajadura-Jiménez
Organizations: School of Computer Science and Creative Technologies, University of the West of England, Bristol, U.K. · Department of Informática, Universidad Carlos III de Madrid, Leganés, Spain. · Architectural Acoustics Research Group, Universidad Politécnica de Madrid, Madrid, Spain. · Department of Signal Theory and Communications, Universidad Carlos III de Madrid, Leganés, Spain. · Department of Media Technology and Interaction Design, KTH Royal Institute of Technology, Stockholm, Sweden. · HCI group, School of Digital Technologies, Tallinn University, Estonia. · Department of Mathematics, Universidad Carlos III de Madrid, Leganés, Spain. · School of Engineering & Informatics, University of Sussex, Brighton, UK. · UCL Interaction Centre, University College London, London, UK.
Multimodal datasets that combine synchronised behavioural and physiological signals with subjective affective and body perception measures are still rare. This limits both the understanding of embodied experience and the development of computational models that infer subjective states from sensor data. We introduce BODYinTRANSIT-Lab, a multimodal dataset that addresses this gap by capturing affective, perceptual, and behavioural responses to the Footsteps Illusion. In this closed-loop paradigm, participants perform walking tasks while receiving real-time manipulated auditory feedback of their own footsteps intended to alter perceived body weight. The dataset comprises synchronised recordings from 101 participants recruited through systematic stratification by eating-disorder symptomatology and physical activity level. BODYinTRANSIT-Lab combines full-body kinematics with electromyography, electrocardiography, respiration, and electrodermal activity. Its annotations capture both immediate and post-trial experience, including within-trial perceived-weight ratings, post-trial assessments of body feelings and affective experience, 3D avatar body visualisations, and quantitative body maps. Psychological, behavioural, demographic, and social measures provide further information about factors that may influence individual responses. We provide a deep learning benchmark that uses walking kinematics to classify relative affective valence and perceived weight within subjects. On within-session, segment-level splits, the baseline model achieved mean balanced accuracies of 45.9% for affective valence and 44.5% for perceived weight. The BODYinTRANSIT-Lab dataset provides an experimentally controlled resource for cognitive neuroscience and affective computing research, with potential applications in affect-aware interactive systems, health and well-being technologies, and creative embodied experiences.
Figures & tables
Fig. 1: Conceptual framework for Body Transformation Experiences (BTEs), linking multisensory input, social and contextual factors, individual characteristics, body representation, and behavioural and affective responses. Arrows depict conceptual relationships and feedback, rather than causal pathways established by the present dataset.
Resource
Experimental variation
Bodily data
Annotations & participant measures
Emilya [ 21 ]
Acted emotional states across seven daily actions
Full-body inertial motion capture; video; audio
Intended emotion and third-party ratings of expressed emotion.
Li et al. [ 31 ]
Walking before and after video-based induction of anger and happiness
Kinect-based 3D joint trajectories
Self-reported anger and happiness around elicitation and walking.
EmoPain [ 3 ]
Rehabilitation activities under normal and increased physical demand
Full-body 3D motion capture; EMG; video; audio
Pain and anxiety reports; distress and catastrophising measures; expert annotations of pain-related expression and behaviour.
EmoPain@Home [ 39 ]
Naturally occurring everyday activities at home
Wearable inertial motion capture
Activity labels and repeated reports of pain, worry, and movement confidence.
PsyMo [ 9 ]
Seven walking variations, including speed, carrying, and dual-task conditions
Video-derived silhouettes; 2D/3D skeletons; 3D body meshes
Personality, self-esteem, fatigue, aggression, and mental-health measures.
Zhao et al. [ 68 ]
Between-person variation in depression severity and personality
Video and audio; ankle-worn IMU gait data
Depression severity and personality.
TABLE I: Selected affective-movement resources compared by experimental variation, recorded or released bodily data, and relevant annotations. Participant-level measures do not imply continuous annotation.
Fig. 2: Left: real-time footstep feedback under control (C), high-frequency (HF), and low-frequency (LF) conditions; curves indicate frequency shaping. Centre: intended perceptual changes relative to the physical body, assessed subjectively. Right: kinematic and physiological acquisition, aligned by accelerometer cross-correlation, with start/stop coordinated by Node-RED [ 40 ] . Sensor locations are schematic.
Fig. 3: Experimental setting and tasks. Left: participant equipment. Centre: walking between A and B with spoken perceived-weight ratings. Right: post-trial questionnaires, avatar adjustment, and body maps. The sensor locations are shown in Fig. 2
Component
Details
Motion capture suit
Rokoko Smartsuit Pro II with 19 IMU sensors, sampled at 100 Hz.
Physiological sensing
BITalino 8-channel biosignalsplux kit; sampling rate 1000 Hz.
Auditory feedback
SoniWeight Shoes (custom) [ 15 ] .
Microphones
Core Sound binaural microphone.
DSP board
Bela.io Starter Kit; average latency 1.6 ms.
Headphones
Sennheiser HDA300; closed-back, circumaural.
TABLE II: Hardware and software specifications used in the experimental setup, including data acquisition and sound playback systems.
Kinematic quantity
Anatomical locations
Groups
Channels
Segment linear kinematics
Position, velocity, acceleration (X,Y,Z)
17 body segments ‡
17 each
153
Joint angular kinematics
Flexion
Hips, knees, shoulders, elbows, wrists, neck
11
33
Abduction
Wrists, elbows, shoulders, hips
6
18
External rotation
Shoulders, hips, knees
6
18
TABLE III: Kinematic channels exported by the Rokoko Smartsuit Pro II. Segment linear kinematics comprise position, velocity, and acceleration along the X , Y , and Z axes. Each joint-angular group comprises angle, angular velocity, and angular acceleration. ‡ 17 body segments include the head, chest, pelvis, shoulders, upper arms, forearms, hands, upper legs, lower legs, and feet.
Signal
Filtering
Filter parameters
Subsequent processing
Respiration
Butterworth IIR band-pass
Order 2; 0.1–0.35 Hz
Adaptive Hampel filtering with window length derived from the detected breathing rate; 3-SD threshold.
EMG
FIR band-pass; Butterworth IIR notch
Band-pass order 20, 20–450 Hz; notch order 2, 49–51 Hz
Hann window followed by adaptive Hampel filtering with a window length derived from the dominant frequency.
TABLE IV: Preprocessing of BITalino signals. Filters were applied using zero-phase forward–backward filtering. Hampel filtering removed transient outliers using fixed or signal-dependent window lengths.
Component
Measures or identifiers
Level and timing
Representation
Participant characteristics and identifier
Participant identifier
Unique participant ID
Participant
Integer identifier
Demographic measures
Age, sex, gender, education
Participant; before session
Numeric or categorical values
Anthropometric measures
Height, waist, hip circumferences (cm); weight (kg); shoe size (EU)
Participant; before session
Numeric values in the stated units
Body-related self-perceptions
Actual and ideal gender-identity ratings; ideal weight (kg)
TABLE V: BODYinTRANSIT-Lab data inventory, grouped by participant characteristics, experimental identifiers and auditory manipulation, sensor recordings, and subjective responses. The table distinguishes each entry by its level and collection timing. Detailed acquisition specifications are given in Table II , and kinematic channels are listed in Table III .
Fig. 4: The network architecture used for the computational benchmark.
Fig. 5: Confusion matrices for relative (a) affective valence and (b) perceived weight. Window-level predictions and targets were pooled across participants and 10 random seeds. Ref. (0) includes control trials and manipulated-feedback trials rated equal to the participant-specific control mean. Negative/positive labels indicate lower/higher valence or lighter/heavier perceived weight relative to that mean.
Task
N/pool
Distribution
Accuracy
BA
F1
> ref.
Auditory condition
99/99
34.4/33.8/31.8
34.8
36.6
28.5
61.6
Relative perceived weight
92/98
23.3/41.4/35.3
43.9
44.5
37.6
81.5
Relative valence
83/98
26.5/47.0/26.5
47.9
45.9
39.4
84.3
TABLE VI: Within-participant classification performance across eligible participants and 10 random seeds. N /pool denotes the number of eligible participants relative to the task-specific pool. Class distributions are reported as average participant-level percentages, with subjective-task class distributions ordered as [−1,0,+1] . > ref. denotes the percentage of participants with balanced accuracy exceeding the nominal three-class reference value of 33.3%. All performance values are percentages.
Fig. 6: Attention profiles for (a) relative affective valence and (b) relative perceived weight. Weights were averaged within participants across test windows and then across 10 seeds. Bars show participant-level means averaged across participants, and error bars show population standard deviations across participants.
Appendix figures & tables1 asset
Supplementary material from the paper’s appendix.
Appendix
Characteristic
Group 1 (n=20)
Group 2 (n=21)
Group 3 (n=41)
Group 4 (n=19)
Overall (N=101)
Age (years)
24.25 (8.72)
25.81 (7.38)
25.10 (6.50)
25.11 (7.85)
25.08 (7.32)
Sex (F/M)
16 / 4
11 / 10
19 / 22
8 / 11
54 / 47
Gender (F/M/NB/O)
16 / 3 / 0 / 1
12 / 9 / 0 / 0
18 / 22 / 1 / 0
8 / 10 / 1 / 0
54 / 44 / 2 / 1
Height (cm)
169.55 (9.03)
168.52 (8.30)
171.32 (8.49)
169.58 (7.55)
170.06 (8.35)
Weight (kg)
65.35 (12.21)
65.76 (12.47)
67.85 (10.61)
62.37 (5.86)
65.89 (10.71)
Ideal Weight (kg)
62.06 (11.82)
62.67 (9.17)
65.95 (9.93)
62.76 (6.43)
63.96 (9.65)
Appendix
TABLE S1: Participant Demographics, Anthropometric, Psychometric, and Social Characteristics (N=101). Sex (F/M) : Sex assigned at birth (Woman / Man); Gender (F/M/NB/O) : Self-reported gender (Feminine / Masculine / Non-binary / Other or Prefer not to say); Education Level (1-5) : Ordinal education level (1=Secondary, 2=Vocational, 3=Bachelor’s, 4=Master’s, 5=PhD); Gender Identity (1-7) : Self-perceived gender identity (1=very masculine, 7=very feminine); Ideal Gender Identity (1-7) : Desired gender identity (1=very masculine, 7=very feminine); EDE-Q : Eating Disorder Examination Questionnaire (Global Score); range 0–6, where 0 indicates no/lowest frequency of eating-disorder-related symptoms and 6 indicates the highest frequency/severity of symptoms; IPAQ : International Physical Activity Questionnaire (total score in MET-min/week); quantifies physical activity based on reported activity during the previous 7 days, expressed in Metabolic Equivalent of Task (MET) minutes per week; MBSRQ : Multidimensional Body-Self Relations Questionnaire (MBSRQ) subscale scores: Importance of Corporality (ISC), Behaviours aimed at maintaining physical shape (COMF), Self-evaluated physical attractiveness (AFA), and Physical appearance care (CAF); range 1–5, where higher scores indicate greater endorsement of the corresponding dimension; Betts’ QMI : Overall Imagery Vividness score from the Betts’ Questionnaire upon Mental Imagery; range 1–7, where a higher numerical score actually indicates lower imagery vividness; SSQ6 : Statistics describing the participant’s self-reported social network, as measured by the Social Support Questionnaire (Sex Ratio, Network Size, Group Count, Average Age of network, Age Difference with participant, and Mean & Standard deviation (SD) of people per questionnaire item); assesses perceived social support, including the number of available supportive people and satisfaction with that support, where higher scores indicate greater perceived availability of social support. Group 1: Low EDE-Q / Low IPAQ, Group 2: High EDE-Q / Low IPAQ, Group 3: Low EDE-Q / High IPAQ, Group 4: High EDE-Q / High IPAQ.
Visual perception of walkability varies substantially across individuals, reflecting differences in personal characteristics, experiences, and preferences. Existing studies, however, often reduce these diverse judgements to aggregated scores, implicitly assuming uniform perception, and commonly rely on vehicle-mounted street-view imagery that does not reflect the pedestrian's visual experience. This paper introduces a dataset of 29,870 walkability ratings from 1,196 respondents, linking sidewalk-view imagery across urban, suburban, and regional Australian environments with individual rater attributes, and proposes the first user-conditioned multimodal deep learning framework for walkability perception, fusing visual features with respondent-level representations. A viewpoint-comparison study shows that sidewalk-view images receive significantly higher walkability ratings than matched street-view images, indicating that imagery source is a substantive design decision in perception surveys. The user-conditioned model improves rank agreement with observed ratings by 65% over an image-only baseline (quadratic weighted kappa 0.47 vs. 0.29), demonstrating that who is evaluating an environment carries predictive indication beyond image content alone. These findings support moving from aggregated, observer-independent walkability scores toward models that represent diverse users, enabling more inclusive assessment of pedestrian environments.
Moloud Damandeh, Meead Saberi
School of Civil and Environmental Engineering, University of New South Wales (UNSW), Sydney, NSW, Australia, and the Research Centre for Integrated Transport Innovation (rCITI)
This Data Descriptor presents a fully open, multi-modal dataset for estimating vertical ground reaction force (vGRF) from consumer-grade Apple Watch sensors with laboratory force plate ground truth. Ten healthy adults aged 26--41 years performed five activities: walking, jogging, running, heel drops, and step drops, while wearing two Apple Watches positioned at the left wrist and waist. The dataset contains 492 validated trials with time-aligned inertial measurement unit (IMU) recordings (approximately 100 Hz) and force plate vGRF (Force_Z, 1000 Hz). The release includes raw and processed time series, trial-level metadata, quality-control flags, and machine-readable data dictionaries. Trial-level matching manifests link recordings across modalities using stable identifiers. Of the 492 validated trials, 395 are triad-complete, containing wrist, waist, and force plate data, enabling cross-sensor analyses and reproducible model evaluation. Dataset quality is characterised through a three-phase cross-sensor plausibility and consistency framework, repeatability analysis of peak vGRF (intraclass correlation coefficient 0.871--0.990), and systematic checks of force ranges and trial completeness. Monte Carlo sensitivity analysis showed that correlation-based validation metrics were robust to single-sample timing perturbations at the IMU sampling resolution. All data are released under CC BY 4.0, with analysis scripts archived alongside the dataset and mirrored on GitHub. This resource supports reproducible research in wearable biomechanics, benchmarking of machine learning models for vGRF estimation, and investigation of sensor placement effects using widely available consumer wearables.
Department of Data Science & Artificial Intelligence and Modelling (DAIM), University of Hull, UK · School of Computer Science, University of Hull, UK · School of Science & Technology, Nottingham Trent University, UK
We introduce the BEATS dataset, a robust compilation of psycho-physiological data collected in the wild from 30 participants while they listened to curated playlists intended to promote well-being. An acronym for Biomarkers, Experience and Affect from Therapeutic Soundscapes, the BEATS dataset aims to support research on wellbeing, music psychology, affective computing and signal processing. More specifically, this dataset includes: (1) sensor data collected during music listening sessions using a research-grade wearable device in free-movement environments; (2) music listening activity records, stimulus metadata, acoustic features, and self-reports associated with the listening sessions (3) individual characteristics, such as demographics, music preference, musical skills, and personality traits; and (4) self-reported psychological measures before and after their experimental participation spanning for a number of weeks. By situating data collection outside of lab settings, the BEATS dataset offers the opportunity to study real-world music-listening scenarios, allowing for a richer exploration of the physiological, subjective and therapeutic effects of music under real-world conditions.
Benjamin Gutierrez Serafin, Tanaya Guha, Fahim Kawsar
University of Glasgow, United Kingdom · OmniBuds Ltd, United Kingdom