Robot Navigation in Risky, Crowded Environments: Understanding Human Preferences

People don't navigate risky environments the way they think they do: The study found that participants' self-assessments of their own risk tolerance and time urgency did not actually correlate with the paths they chose in the grocery shopping scenario, suggesting humans may not have accurate insight into their own decision-making in stressful situations.

Human preferences for robot navigation are diverse and unpredictable: Rather than following a single "optimal" path, participants showed a wide variety of preferences ranging from "risky and urgent to safe and relaxed," indicating that robots need to be designed with flexibility to accommodate different human values rather than assuming one-size-fits-all navigation strategies.

Cumulative Prospect Theory better explains human choices than simpler models: When comparing three risk assessment models, Cumulative Prospect Theory (CPT) most accurately captured how people actually made decisions in the risky grocery shopping scenario, outperforming simpler expected risk models—suggesting AI systems should use more sophisticated behavioral models to predict human preferences.

Robots need explainable AI to navigate crowded, uncertain environments responsibly: The research emphasizes that robots deployed in real-world risky and crowded settings require "explainable AI" designs that can interpret human behavior and communicate their navigation choices, making them more trustworthy and safer for public deployment.