CAREER: Toward Guaranteed Reliability in AI-Enabled Robotic Teleoperation Through Formal Logic and Certification

This project will create a safe and reliable way for people and robots to work together through teleoperation (remote control of machines), especially in high-risk settings such as surgery, disaster response, and space or nuclear operations. In these environments, robots must perform precise tasks while responding to human guidance. However, many current artificial intelligence approaches are difficult to understand and verify, which limits trust and wider use in safety-critical applications. This project will develop new methods that will help robots better understand human intent, provide timely assistance, and maintain safe and predictable performance, even in uncertain situations, allowing humans and robots to share control more effectively while reducing workload. The broader impacts of this project will include more trustworthy and reliable robotic systems for healthcare, manufacturing, and exploration in space and the ocean. The project will also promote education in science, technology, engineering, and mathematics by providing hands-on research opportunities and training for undergraduate and graduate students, engaging K–12 students in interactive robotics activities, and collaborating with national laboratories and industry partners to establish safety standards and encourage the responsible use of advanced robotic systems. This project seeks to develop a unified framework for safe, reliable, and interpretable teleoperation, advancing the collaboration between humans and robots in high-risk, unstructured environments. The research integrates AI foundation models with formal methods for safety specification, enabling robotic systems to interpret human intent, adapt to dynamic conditions, and maintain predictable behavior even under uncertainty. Unlike existing teleoperation approaches, which purely rely on opaque AI “black-box” models or oversimplified safety assumptions, this work focuses on producing certifiable and generalizable methods that move beyond basic collision avoidance to address real-world complex safety requirements. The proposed research aims to advance three core capabilities. First, it will develop learning-based assistive primitives that connect operator intentions and generate reusable, specification-compliant actions with minimal training. Second, it will establish logic-guided motion prediction and control frameworks that incorporate uncertainty quantification to ensure safe human-robot collaboration. Third, it will implement real-time monitoring and safety mechanisms capable of detecting anomalies, generalizing to novel conditions, and maintaining situational awareness throughout teleoperation. Collectively, these innovations are designed to enable adaptive, trustworthy, and high-performance human-robot interaction in real-world scenarios, laying the foundation for broader deployment of teleoperated systems. 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. NSF Award ID: 2541947 | Program: 01002627DB NSF RESEARCH & RELATED ACTIVIT | Principal Investigator: MINGYU CAI | Institution: University of California-Riverside, RIVERSIDE, CA | Award Amount: $573,377 View on NSF Award Search: https://www.nsf.gov/awardsearch/show-award/?AWD_ID=2541947 View on Research.gov: https://www.research.gov/awardapi-service/v1/awards/2541947.html