A Dataset of Human-Controlled Tower-Crane Operation: Kinematic Time Series and Event Logs Across Two Viewing Interfaces and Three Task Difficulty Levels
Description
This dataset contains multivariate time-series logs and timestamped operational events collected during human-controlled tower-crane operation in a desktop simulator. It is intended to support research on short-horizon load-sway prediction, temporal modelling of crane movements, and predictive monitoring during simulated material-handling tasks. Data were collected from 18 participants using two viewing interfaces: a first-person cabin view (FPV) and a freely navigable third-person external view (TPV). Each participant completed three task difficulty levels under both viewing conditions, yielding 108 experimental trials. The easy level involved transporting one cargo, while the medium and hard levels each involved transporting three cargos. Viewing order was counterbalanced: nine participants completed FPV first, and nine completed TPV first. The continuous records include timestamps, trolley position, jib rotation, cable length, three-dimensional hook position, and oscillation about two axes. Discrete event records contain event timestamps, event types, and spatial coordinates. Recorded events include trial start and end, cargo pickup and delivery, near misses, and collisions. Continuous measurements and events use separate timestamp fields and should be aligned by time rather than assumed to correspond within the same CSV row. The data are organized by anonymized participant identifiers (P01–P18), with six trial CSV files per participant corresponding to the two viewing conditions and three difficulty levels. Participant, viewing-condition, and difficulty identifiers support comparisons across operating conditions and the construction of participant-independent training and evaluation sets. Pickup and delivery events enable identification of loaded-transport intervals for targeted analysis of load motion. The primary intended reuse is predicting future load oscillation from preceding crane-state and oscillation histories. Additional uses include trajectory analysis, operational-phase segmentation, and investigation of model robustness across users and viewing conditions. For machine-learning applications, training and evaluation sets should be separated by participant before generating temporal windows to reduce information leakage. The records describe human operation within a simulated environment. They are not measurements from a physical tower crane, and models developed using this dataset require independent validation before application to real machinery. The dataset originates from a study approved by Nazarbayev University’s ethics committee, with informed consent obtained from participants.
Files
Steps to reproduce
1. **Simulator development:** A desktop tower-crane simulator was developed in Unity with controls for crane movement and cargo handling. Two viewing interfaces were implemented: a first-person cabin view (FPV) and a freely navigable third-person external view (TPV). 2. **Task preparation:** Three difficulty levels were created. The easy level required transporting one cargo, while the medium and hard levels each required transporting three cargos to designated delivery locations. 3. **Logging implementation:** The simulator was configured to record timestamped trolley position, jib rotation, cable length, three-dimensional hook position, and oscillation about two axes. A separate event stream recorded trial start and end, cargo pickup and delivery, near misses, and collisions, including event timestamps and coordinates. 4. **Participant recruitment and consent:** Eighteen participants were recruited from Nazarbayev University. Following institutional ethics approval, participants received an explanation of the procedure, provided informed consent, and completed a demographic questionnaire. 5. **Familiarization:** Participants received verbal and written instructions on the controls, viewing interfaces, and tasks. They completed one practice trial at each difficulty level under both viewing conditions, with additional practice provided when needed. 6. **Experimental protocol:** Each participant completed both viewing conditions. Viewing order was counterbalanced: nine participants performed FPV first and nine performed TPV first. Within each condition, tasks followed the easy, medium, and hard sequence. 7. **Data collection:** Continuous measurements and operational events were recorded automatically during each experimental trial. Each participant completed six trials, producing 108 trial CSV files across the study. Experimental records were organized into participant folders labelled P01–P18. 8. **Data preparation:** Trial intervals were identified using the “Game Started” and “Game Ended” event timestamps. Continuous measurements and events were aligned using their respective timestamps, and cargo pickup–delivery pairs were used to identify loaded-transport intervals. Any cleaning or derived-data processing should be documented separately from the original recordings.
Institutions
- Nazarbayev UniversityAstana, Astana