Physio-HRC 1.0: A Multimodal Dataset on Cognitive Load and Trust across Different Human-Robot Collaboration Scenarios

Physio-HRC 1.0 is an open, multimodal dataset that connects physiological signals with measures of mental workload and trust during human–robot interaction in teleoperation tasks. The dataset was collected from 112 participants across four experimental studies involving: Guiding a simulated quadruped robot (Unitree Go2) across terrains of varying difficulty in NVIDIA Isaac Gym Controlling a UR5 robotic arm mounted on a Husky mobile platform with either limited or assisted camera views Repeating the same task while introducing robot errors at varying autonomy levels Remotely operating a Franka Panda arm from another country using a joystick, touchscreen interface, or one-click autonomous mode Throughout each trial, multiple data streams were recorded: Eye-blink activity (MediaPipe Face Mesh, ~10 Hz) Facial temperature — nose, cheeks, forehead ROIs (Optris PI 640i thermal camera) Skin conductance (GSR) and PPG (Shimmer3 GSR+, 128 Hz) NASA-TLX workload self-reports Trust Perception Scale-HRI questionnaires Task performance metrics (success rate, completion time, task-specific measures) Five experiments are included: ExperimentRobotNConditions 1. Mobile Manipulation with AI GuidanceUR5 on Husky183 guidance levels 2. Cross-Border TeleoperationFranka Panda223 interface types 3. Legged Robot SimulationUnitree Go2254 terrains × 2 stages 4. Autonomy & System ErrorsUR5 on Husky193 autonomy/error levels 5. Information Loss & GuidanceUR5 on Husky283 feedback levels All data are fully de-identified following GDPR. Ethics approvals from Bristol Robotics Laboratory (Ref. BRL-25-04) and ENSTA Paris. See the included README.md, dataset_metadata.json, and sensor_metadata.json for full documentation.