Develops modular digital twins for autonomous accelerator and robotics control, combining hybrid AI and LLM interfaces to enable interpretable, adaptive, and energy-efficient operations.
MODULARITY (Modular Digital Twins and Hybrid AI for Autonomous Accelerator and Robotics Optimization) aims to redefine AI-driven control systems through hierarchical multi-agent architectures and human-in-the-loop intelligence.
Objectives
- Design a modular twin framework with local (ground) and global (intermediate) agents.
- Enable hybrid, interpretable optimization using RL, Bayesian, and MARL methods.
- Integrate LLM-based natural language interfaces for operator transparency.
- Bridge the model-reality gap through continuous learning and adaptive correction.
- Promote sustainability, minimizing downtime and energy consumption.
Innovation Beyond State of the Art
MODULARITY introduces a three-layer AI architecture that unifies control, interpretability, and autonomy:
- Ground agents for subsystem-level optimization and self-healing.
- Intermediate agents for global coordination, drift detection, and health monitoring.
- LLM operator layer ensuring explainability and seamless human collaboration.
It pioneers adaptive hybrid optimization, ensuring real-time learning and scalability across complex physical systems.
Expected Impacts
- Scientific: Next-generation AI control frameworks for accelerators and robotics; improved modeling fidelity and reproducibility.
- Technological & Industrial: Scalable twin infrastructure for autonomous manufacturing and scientific facilities.
- Environmental: 15–25% reduction in energy consumption through proactive, AI-driven optimization.
- Societal: Enhanced safety, reduced downtime, and democratized access to advanced control systems.
Supporting Institutions
This initiative brings together key European players in accelerator physics, robotics, and AI research