CAREER: Enhancing the Robustness of Human-Robot Interactions via Automatic Scenario Generation

University Of Southern California, Los Angeles CA

Investigators

Abstract

There is a need for new techniques to assess humans and robots’ interactions at home and in the workplace. Traditionally, human-robot interaction is tested with human subject experiments. While these experiments are necessary to evaluating human-robot interactions, they are often limited in the number of environments and human behaviors that can be observed. Furthermore, it is not well understood how to build robots that account for infrequent and undesirable behaviors found when testing such systems. This Faculty Early Career Development (CAREER) award supports fundamental research to improve human-robot interactions by automatically creating simulated human-robot interaction scenarios that reveal undesirable behaviors, as well as integrating the generated scenarios into the robot’s learning process. Results from this work will provide the field of robotics with a theoretical and experimental tools for allowing the robot to adjust to new and challenging scenarios. Tightly integrated with the research activities, the education plan will introduce scenario generation in robotics education and artificial intelligence competitions to improve students' understanding of robots' capabilities and limitations and inspire them to pursue a career in science, technology, engineering, and mathematics. This project will advance the science of robust, complex human-robot interaction by automatically generating and learning from diverse, challenging and realistic scenarios in simulation. It will investigate computational foundations for the design of quality diversity algorithms that efficiently search the scenario space. It will then develop frameworks that integrate the developed algorithms with generative models to optimize a low-dimensional space of complex and realistic scenarios. The project will close the loop between scenario generation and learning by exploring and characterizing methods for efficiently selecting challenging scenarios to form a curriculum for learning. Dissemination of all developed algorithms through open-source software and workshops will help bring ideas from quality diversity optimization and scenario generation to a wider robotics audience. This project is supported by the cross-directorate Foundational Research in Robotics program, jointly managed and funded by the Directorates for Engineering (ENG) and Computer and Information Science and Engineering (CISE). This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.