2020/12/08 by Matthew C. Fontaine, Stefanos Nikolaidis, Fontaine, Matthew +1 · 1 citation
Computer Science · Psychology · #FOS: Computer and information sciences #Human-Automation Interaction and Safety #Reinforcement Learning in Robotics #Robotics (cs.RO) #Software Reliability and Analysis Research
paper · pdf · doi:10.48550/arxiv.2012.04283
openalex publication_date 2020/12/08 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
The growth of scale and complexity of interactions between humans and robots\nhighlights the need for new computational methods to automatically evaluate\nnovel algorithms and applications. Exploring diverse scenarios of humans and\nrobots interacting in simulation can improve understanding of the robotic\nsystem and avoid potentially costly failures in real-world settings. We\nformulate this problem as a quality diversity (QD) problem, where the goal is\nto discover diverse failure scenarios by simultaneously exploring both\nenvironments and human actions. We focus on the shared autonomy domain, where\nthe robot attempts to infer the goal of a human operator, and adopt the QD\nalgorithm MAP-Elites to generate scenarios for two published algorithms in this\ndomain: shared autonomy via hindsight optimization and linear policy blending.\nSome of the generated scenarios confirm previous theoretical findings, while\nothers are surprising and bring about a new understanding of state-of-the-art\nimplementations. Our experiments show that MAP-Elites outperforms Monte-Carlo\nsimulation and optimization based methods in effectively searching the scenario\nspace, highlighting its promise for automatic evaluation of algorithms in\nhuman-robot interaction.\n