CPS-FR: Toward Cyber-Physical-Cognitive Collaborative Systems for Epistemic Human-Robot Teaming

Robots are increasingly being deployed in homes, hospitals, factories, warehouses, transportation systems, and disaster response operations. Although modern robots can perform many complex tasks, they still struggle to collaborate naturally with people because they lack a fundamental human capability: the ability to reason about what others know, intend to do, and expect from their partners. Humans routinely make these inferences when working together, allowing teams to coordinate efficiently with little communication. Enabling robots to reason in similar ways is one of the central scientific challenges in robotics and artificial intelligence. This project establishes the scientific foundations for a new generation of Cyber-Physical-Cognitive systems that combine physical interaction, intelligent decision making, and cognitive reasoning about human partners. Rather than simply reacting to human actions, robots developed through this research will infer people's goals, estimate what they believe about the environment, anticipate what they expect the robot to do, and adapt their own behavior accordingly. These capabilities will enable robots to provide proactive assistance, communicate more effectively through their actions, and collaborate safely even when communication is incomplete or unavailable. To achieve these goals, the project advances fundamental knowledge at the intersection of robotics, artificial intelligence, cognitive science, and cyber-physical systems by introducing new principles for integrating perception, reasoning, planning, learning, and control. Broader impacts of this work extend across many sectors that increasingly rely on human-autonomy teaming, including advanced manufacturing, healthcare, transportation, emergency response, infrastructure inspection, and national defense. The project has the potential to improve safety, productivity, resilience, and quality of life while reducing cognitive workload for human participants. The project also contributes to education and workforce development by training undergraduate and graduate students in robotics, artificial intelligence, and cyber-physical systems; integrating the research into new educational materials and university courses; engaging K-12 students through interactive demonstrations that illustrate human-robot teamwork and scientific reasoning; and releasing open-source software that enables researchers, educators, and students to build upon the project's results. Together, these activities strengthen the nation's scientific leadership in intelligent autonomous systems while preparing the next generation of engineers and researchers. The central scientific premise of the project is that effective human-robot collaboration requires robots to model not only what humans intend to do, but also what humans believe the robot intends to do. This recursive reasoning, inspired by the cognitive concept of Theory of Mind, provides a foundation for autonomous systems that coordinate naturally with people and enables a robot to anticipate mental states of team members. The project is organized in three tightly integrated research thrusts. The first introduces an epistemic task and motion planning framework that represents and propagates multiple hypotheses about human beliefs and expectations while jointly planning collaborative robot actions. The planner combines probabilistic belief reasoning, active inference, and Monte Carlo Tree Search to generate actions that simultaneously advance shared objectives, reduce uncertainty, and improve collaboration. To ensure that these decisions remain both safe and understandable to human partners, the project introduces Epistemic Control Barrier Functions, a new control-theoretic framework that embeds cognitive reasoning directly into continuous robot control while providing formal guarantees on safety and behavioral consistency. The second thrust develops methods for social epistemic inference, enabling robots to infer human goals, beliefs, and future actions from multimodal observations. The proposed neuro-symbolic framework will allow to infer human intent while grounding all inferences in observed behavior. These estimates are coupled with multimodal human motion prediction models that capture diverse human strategies and enable personalized human-robot interactions. The third thrust develops robust epistemic planning methods that account for uncertainty in perception, intent estimation, and human behavior. The project combines adaptive dynamical models, online personalization, active information gathering to quantify uncertainty and provide statistically grounded guarantees for safe decision-making during human-robot interaction. Collectively, these contributions establish new theoretical foundations for cognitive autonomy by integrating perception, reasoning, planning, learning, and control within a unified Cyber-Physical-Cognitive architecture. 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.