This position paper presents a vision for self-driving particle accelerators that operate autonomously with minimal human intervention. We propose that future facilities be designed through artificial intelligence (AI) co-design, where AI jointly optimizes the accelerator lattice, diagnostics, and science application from inception to maximize performance while enabling autonomous operation. Rather than retrofitting AI onto human-centric systems, we envision facilities designed from the ground up as AI-native platforms. We outline nine critical research thrusts spanning agentic control architectures, knowledge integration, adaptive learning, digital twins, health monitoring, safety frameworks, modular hardware design, multimodal data fusion, and cross-domain collaboration. This roadmap aims to guide the accelerator community toward a future where AI-driven design and operation deliver unprecedented science output and reliability.
Imagine a particle accelerator that runs itself, automatically tuning thousands of magnets and RF cavities for optimal beam quality, detecting and diagnosing equipment faults before they cause downtime, and adapting in real-time to changing experimental demands --all with minimal human intervention. As accelerators grow even more powerful and intricate, with millions of sensor channels and thousands of interconnected components that must be precisely coordinated, human operators will be stretched to their limits. This has prompted a fundamental question: can we operate accelerators more autonomously, with AI managing complexity at machine speed while humans provide strategic oversight?
The vision of a “self-driving, natively-AI” particle accelerator is a facility that runs optimally with minimal human intervention, continuously tuning itself, diagnosing issues, and safely adapting to changing conditions in real time. Realizing this vision requires rethinking both control systems and accelerator design from the ground up. An autonomous accelerator needs AI systems deeply integrated at every level, from low-level device control up to high-level decisionmaking. Crucially, the accelerator must be natively AI-driven, meaning it is engineered for autonomy from the outset rather than having AI added as an afterthought. Every subsystem should be fully instrumented, digitally accessible, and outfitted with automation hooks for AI to auto-configure, auto-stabilize, auto-analyze, and auto-recover as needed. Like a driverless car’s sensor fusion and autopilot layers Kocic et al. [2018], future accelerators will embed intelligent diagnostics and control loops throughout for safe and efficient operation. The following sections outline this vision in more detail and draw connections to current advances that hint at what is possible Edelen and Huang [2024]. This vision aligns directly with recent national priorities, as evidenced by the Department of Energy’s designation of “Enhancing Particle Accelerators for Discovery” as one of 26 Genesis Mission Science and Technology Challenges, specifically calling for deploying AI to make accelerators adaptive and autonomous to accelerate breakthroughs in medicine, materials, and energy U.S. Department of Energy [2026].
Enabling a truly autonomous accelerator calls for a paradigm shift in how we design and operate these machines. In this paradigm, AI agents serve as the primary operators of the facility, while humans shift to supervisory and strategic roles. The accelerator complex would effectively become a cyber-physical AI system. Every cavity, magnet, diagnostic, and power supply is not only digitally controlled but also paired with AI routines that monitor and adjust its performance. On-site human intervention would become a last resort. Achieving this means engineering accelerators with autonomy as a core design goal, incorporating built-in safety margins, high component reliability, modular designs, and redundancy so that automated adjustments or component swaps can occur without jeopardizing operations.
The vision of an autonomous, natively-AI accelerator begins not with operations, but with design. A truly autonomous facility must be conceived through AI co-design from its inception, where AI shapes both the accelerator architecture and its science application as a unified, jointly-optimized system. This represents a fundamental departure from conventional approaches where accelerator design and experimental requirements are treated as separate optimization problems, connected only through interface specifications and performance requirements. AI-driven design would leverage accelerator-specific knowledge bases or foundation models to probe vast solution spaces and identify novel lattice configurations that go beyond current state-of-the-art performance Ji et al. [2024]. Rather than relying solely on human intuition and incremental improvements to established designs, AI systems can explore unconventional combinations of components to discover configurations that beam physicists might never consider. Crucially, this optimization extends beyond the traditional accelerator components to include the diagnostics themselves -which sensors to deploy, where to place them, and how to configure them for optimal self-diagnosing capabilities. An AI-designed accelerator would thus be born with the instrumentation and observability required for autonomous operation, rather than having diagnostics added as an afterthought.
More significantly, the science application -whether it be a nuclear physics experiment or a materials characterization beamline -would be jointly optimized with the accelerator lattice from the very beginning. AI co-design enables optimization of the entire system -accelerator and experiment together -toward maximizing specific science objectives. This integrated approach promises to generate outsized science output compared to conventionally designed facilities.
This content is AI-processed based on open access ArXiv data.