DU Want To Build-A-Bot?

A common issue encountered by practitioners of Artificial Intelligence (AI) and Machine Learning (ML) is a lack of salient data to use in training. A common issue in Human-Robot Interaction is a gap in understanding how robot designs are perceived by the user. The "D U Want to Build A Bot" (Build-A-Bot) project is developing a novel robotic design research platform implemented as a web-accessible 3D game that will allow us to quickly gather many user-provided robot design examples. These examples are then used to train ML models to better evaluate robot designs, predict how a design will be perceived, and create new robot designs. It is anticipated that we use Convolutional Neural Networks (CNNs) to predict how an existing robotic design will be perceived, and Generative Adversarial Networks (GANs) to create new robot designs based on the user provided examples. Additionally, neuroscience methods including functional near-infrared spectroscopy (fNIRS) will be explored as strategies for validating the created ML models.