Quantifying Human Trust in Controlling Robot Swarms: EEG-based Analysis and Classification

Trust is a cornerstone of human-to-human inter-action. We build up or tear down relationships depending on our perceived trust levels. However, we tend to be cautious when trusting artificial systems such as robots or artificial intelligence. This article addresses the gap in determining levels of trust in scenarios where humans interact with robot swarms by directly providing command inputs. We use Electroencephalography (EEG) data of human subjects controlling robotics swarms via joystick commands and randomly inject enough noise to make their given task difficult. In doing so, we artificially create distrust in the system and use machine learning techniques to find EEG correlates of swarm trust. The results of this study suggest that, when directly interacting with a robot swarm, these neural correlates of trust in a robot swarm exist and are discernable via machine learning feature classification to an accuracy greater than 88%. This work shows great promise in establishing effective human-machine teaming in exploration, search and rescue operations, and defense. Successfully quantifying human trust levels is an essential building block to facilitate the adoption of robot swarms in real-world environments with human operators, as swarms can leverage these to adapt their behavior to maintain or regain a human's trust level in the team.