Experimental environment used for human-robot coexistence data collection

MULTIMODAL ROBOTICS DATASET

Human-Robot
Coexistence Dataset

High-precision multimodal recordings of humans and mobile robots sharing dynamic environments.

32 Experimental
Trials
3.7h Total
Recording Time
142 Human
Tracks
2 Mobile
Robots

A dataset for understanding human-robot coexistence.

CAGE contains synchronized multimodal recordings of humans and mobile robots sharing a controlled dynamic environment.

The dataset combines motion-capture trajectories with onboard robot sensor recordings, time-stamped pedestrian goals, experimental context, and subjective survey responses.

It is designed to support research in human-robot interaction, navigation, coordination, intent communication, and socially aware autonomy.

View the project on GitHub →

From the physical setup to synchronized data.

CAGE was collected in a controlled indoor environment where human participants and mobile robots performed goal-directed movements under different interaction conditions.

01

Experimental environment

Experiments were conducted in a controlled 7 × 6 m indoor space equipped with a ceiling-mounted OptiTrack motion capture system. Human participants and mobile robots shared the same physical environment while their movements were recorded continuously.

The motion-capture system provided continuous 6-DoF pose measurements for tracked humans and robots throughout the experiments.

7 × 6 m Experimental space
24 OptiTrack cameras
120 Hz Tracking frequency
< 0.35 mm Average calibration residual
Experimental environment used for CAGE data collection
Controlled indoor environment used for human-robot coexistence experiments.
Map of the experimental environment
Layout of the experimental space and navigation area.
02

Experimental layout

The experimental area was configured to provide a shared space for human movement and robot navigation. The controlled layout allows trajectories and interactions to be compared across experimental conditions.

03

Human tracking

Participants wore uniquely identifiable tracking helmets equipped with OptiTrack markers. This enabled continuous tracking of individual participants throughout each trial.

Participant identities were linked with experimental goals, timestamps, and scenario metadata, allowing human trajectories to be analysed together with task and interaction context.

Tracking helmet worn by a participant
Unique tracking helmets used to identify participants.
04

Mobile robots

Two TurtleBot 4 Lite platforms were used to introduce autonomous robot motion into the shared environment.

The robots represented two different navigation behaviours, providing complementary sources of robot motion and interaction data.

Base TurtleBot 4 Lite Base Random-goal navigation
Delivery TurtleBot 4 Lite Delivery Cyclic multi-goal navigation
Human goal generation interface Goal generation system interface
Goal-generation system used to assign participant goals during experiments.
05

Human goal generation

Human participants received movement goals through a dedicated goal-generation system. The experimenter selected the participant's tracking ID and generated the next goal during the trial.

Each generated goal was recorded together with its timestamp, providing explicit task annotations that can be aligned with the corresponding human trajectory.

06

Ground-projected robot intent

In the explicit-intent condition, a ceiling-mounted hardware-in-the-loop projection system displayed the robot's planned goal and path directly on the floor.

This provided a visible representation of the robot's intended movement and enabled controlled comparison between implicit and explicit robot intent communication.

Hardware-in-the-loop floor projection used to communicate robot intent
Hardware-in-the-loop system used for ground-projected robot goals and paths.
07

Recorded trajectories

The dataset provides synchronized human and robot trajectories, enabling analysis of motion patterns, interactions, and robot-human spatial relationships throughout the experiments.

Human and robot trajectories

Four controlled conditions.

The dataset spans four experimental conditions designed to separate environmental effects, robot presence, and explicit robot intent communication.

01

Human movement

Participants move through the environment without active robots, providing a baseline for human motion and task behaviour.

02

Environmental obstacles

Human movement is recorded in the presence of environmental obstacles without active robot interaction.

03

Implicit robot intent

Two mobile robots operate in the shared environment without explicit visual communication of their intended motion.

Modalities

CAGE connects precise motion measurements with task, robot, experimental, and subjective information.

01

OptiTrack

High-precision 6-DoF trajectories for human participants and mobile robots.

02

ROS 2 Bags

Robot telemetry and onboard sensor recordings collected during selected trials.

03

Human Goals

Goal locations and timestamps generated for participants during each trial.

04

Survey Data

Participant responses measuring perceptions and preferences across experimental conditions.

Use the dataset.

01

Access

The complete CAGE dataset will be published on Zenodo. Until the public release, contact the authors for access to the current research release.

Contact the authors →
02

Explore

Inspect trajectories, human goals, robot data, scenarios, surveys, and experimental context.

03

Analyse

Use the provided toolkit to load, process, visualize, and analyse the recordings.

Cite CAGE in your research.

The CAGE dataset citation will be provided with the public Zenodo release and associated publication.

For access to the current dataset or citation information, please contact the authors.

Contact the authors →
Finnish Center for Artificial Intelligence Aalto University

This work was conducted at Aalto University and acknowledges the support of the Microelectronics and Digital and Autonomous Systems (MIDAS) infrastructure.