Ethically Responsible Human-Robot Interaction

As part of my coursework at Carnegie Mellon University Africa, I collaborated on a project titled “Ethically Responsible Human-Robot Interaction (ER-HRI).” This research explored how humanoid robots can dynamically evaluate ethical dilemmas using machine learning models, integrating ethical frameworks into their decision-making processes. The project focused on using datasets based on justice and common-sense ethics to train these models. We aimed to create a robot capable of navigating complex ethical challenges that arise during human-robot interactions.

Our approach involved building two models—Random Forest and GPT-2—and training them to classify ethical and non-ethical scenarios. We utilized a combined dataset from the ETHICS dataset to train these models. The GPT-2 model outperformed the Random Forest model with an accuracy of 73.3% compared to 56.0%, highlighting the power of advanced natural language models in understanding ethical distinctions.

To implement the models in a virtual environment, we tested the Random Forest model in the Unity platform using a humanoid character, where the robot performed actions based on the model’s ethical judgments. The project also included the assembly of Eva, a humanoid robot with facial expressions, to serve as a real-world testbed for ethical decision-making.

One of the key challenges in this project was assembling Eva, including 3D printing parts and incorporating sensors for human detection. We further integrated OpenAI’s ChatGPT-3 into Eva to simulate ethical decision-making, and though we faced challenges with deployment, this project demonstrated significant progress toward real-world implementation of ethical humanoid robots.

This work has contributed to advancing ethically responsible interaction frameworks in robotics, aiming to ensure that future robots can operate in diverse environments with a strong ethical foundation.

Video of the Build Process

Download the final Report