Collaborative Research: CPS-FR: Closed-Loop Digital Twins for Adaptive Human-AI Collaboration in Cyber-Physical Systems

Modern manufacturing environments increasingly rely on collaboration between human workers, robots, and intelligent software systems to support flexible, high-throughput production. However, current artificial intelligence (AI) systems remain largely reactive and brittle, lacking the ability to understand human intent, anticipate procedural deviations, and adapt to changing task conditions in real time. Even small errors in task execution can propagate through assembly workflows, leading to costly rework, delays, safety risks, and reduced productivity. This project seeks to develop adaptive cyber-physical systems that proactively support human workers via closed-loop digital twins, intelligent models that monitor task execution, reason about procedural intent, and deliver real-time corrective feedback. By enabling intelligent systems to function as collaborative partners rather than passive automation tools, the project advances safer, more resilient, and more efficient manufacturing systems that strengthen U.S. manufacturing competitiveness and workforce readiness. In addition to its scientific contributions, the project supports education and workforce development through hands-on training in cyber-physical systems, robotics, and AI-enabled manufacturing using Auburn University's Tiger Motors manufacturing testbed. This project integrates multi-view event perception, neuro-symbolic reasoning, causal task modeling, and simulation-guided verification for adaptive human-AI collaboration in manufacturing cyber-physical systems. The specific research objectives are four-fold. First, the project seeks to develop a neuro-symbolic framework for real-time event perception and procedural grounding that enables intelligent agents to infer task progress and human intent from streaming sensor observations. Second, the project investigates causal task modeling using predictive coding and Monte Carlo Tree Search to reason over long-horizon task dependencies, forecast execution errors, and generate corrective interventions before failures propagate downstream. Third, the project develops simulation-guided digital twin models that leverage generative visual reasoning and visual perspective taking to validate predicted task trajectories and assess their feasibility in dynamic assembly environments. Finally, the project realizes these capabilities within an AI-ready collaborative manufacturing testbed that enables closed-loop evaluation of intelligent poka-yoke agents in realistic human-robot assembly settings. Together, these contributions advance adaptive, perception-driven cyber-physical systems that tightly couple sensing, reasoning, simulation, and human collaboration in complex real-world environments. 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.