Project

Human-AI Trust Simulation Testbed

A Robot Operating System (ROS)/Gazebo testbed for studying how people calibrate trust in an autonomous robotic teammate during multi-step tasks.

Overview

Controlled human-robot interaction for trust research.

The testbed places a human operator and a simulated TurtleBot in shared task scenarios. The robot interprets human input, completes objectives, and records task outcomes for trust and collaboration research.

The system was built as a public-safe version of the original experiment package, with participant data and runtime state removed.

Citation

IEEE CogMI 2024 Best Paper

Frericks, J., Kang, B., Outland, N., Doshi, P., Johnsen, K., & Schecter, A. (2024). Trust and collaboration testing in controlled human-robot environments. In 2024 IEEE 6th International Conference on Cognitive Machine Intelligence (CogMI) (pp. 127-136). https://doi.org/10.1109/CogMI62246.2024.00026

Demo

Testbed video

Full demo video of the Human-AI Trust Testbed package.

Visuals

System diagrams and experiment views

DAC testbed system diagram
Experiment flow and system structure.
Human robot interaction diagram
Human-robot interaction setup.
Experiment environment layout
Environment layout.

System

What the testbed includes

Stack

ROS, Gazebo, and RViz

Implemented as a ROS catkin package with Gazebo simulation, RViz views, navigation workflows, and Python ROS nodes.

Interface

Human-facing tasks

PyQt interfaces present task prompts, QR and puzzle clues, maze navigation, wire identification, and operator guidance steps.

Model

Trust research workflow

Task outcomes and interaction events are captured as the human-robot team progresses through the experiment.

Environment

Gazebo scenes and task prompts

Overhead Gazebo environment view
Overhead Gazebo environment with task spaces.
TurtleBot interaction scene in Gazebo
Simulated TurtleBot and human-facing task context.
Maze and tripwire task
Maze and tripwire task prompt.
Wire identification task example
Wire identification task.
Second wire identification task example
Additional wire task prompt.
Puzzle clue example
Puzzle clue shown to the operator.
Puzzle target example
Puzzle target example.
Card clue image
Image-based clue used in the task flow.

Impact

Publication and recognition

Paper

Trust and Collaboration Testing in Controlled Human-Robot Environments

Published at IEEE CogMI 2024 and recognized with a Best Paper Award.

Study

DAC-funded experiment package

The testbed supported a funded study of trust and collaboration in controlled human-robot environments.

Links