A study by mechanical engineers and computer scientists at the University of California San Diego explored human risk aversion when interacting with robots in everyday settings. Presented at the ICRA 2024 conference in Japan, the research is said to be the first to investigate robots that infer human perception of risk for intelligent decision-making.
The study, led by Aamodh Suresh, a postdoctoral researcher at the U.S. Army Research Lab, and Angelique Taylor, a faculty member at Cornell Tech, aimed to develop a framework to understand human risk aversion in robot interactions. It took place during the pandemic, necessitating an online experiment in which subjects, mainly STEM students, navigated a virtual grocery store as Instacart shoppers. They chose between paths of varying lengths and COVID-19 exposure risks to reach the milk aisle, with the shortest paths involving the highest exposure risk but offering rewards for quicker completion.
Results indicated that participants underestimated their willingness to take risks in survey responses compared to their actual behavior in the game, particularly when a reward was involved. This led researchers to adopt prospect theory, a behavioral economics model by Nobel laureate Daniel Kahneman. The theory posits that people weigh losses and gains against a reference point and feel losses more intensely than gains. In the study, participants preferred the certain reward of completing the task quickly over the potential risk of COVID-19 exposure.
Researchers also gathered preferences on how participants would like robots to communicate intentions, with suggestions including speech, gestures, and touch screens. The team plans to conduct an in-person study with a more diverse participant group in the future.
Image credit: University of California San Diego
