A Hybrid Dataset for Studying Human Trust Dynamics in Sequential Human-Robot Collaboration

This dataset, A Dataset for Trust Dynamics in Human–Robot Collaboration, provides multi-modal recordings and annotations of human trust evolution during sequential collaborative tasks. The data were collected through a unified experimental framework combining three modalities: (1) virtual human-in-the-loop experiments in VR environments, (2) large language model (LLM)–based human simulation, and (3) real-world quadruped robot collaboration experiments.

Each trial captures the temporal evolution of human trust alongside task state, observations, robot recommendations, human decisions, and rewards, forming a complete trajectory of trust dynamics. The dataset includes 10 sequential subtasks per trial.

Key variables include:

real_state — true environment state per timestep.

observation — human observation under uncertainty.

trust — normalized human trust value .

robot_action — decisions made by robot.

human_action — decisions made by human.

reward — task feedback.

The dataset is designed for research in trust prediction, human–robot collaboration modeling, and trust-aware reinforcement learning.All data are anonymized and formatted as JSON and CSV files for easy processing. A detailed README and schema description are included in the release package.

Usage Notes:Researchers are encouraged to use this dataset for model development, benchmarking, and evaluation of trust-aware AI systems. Please cite this dataset as indicated below when used in publications.