Agentic Physical AI toward a Domain-Specific Foundation Model for Nuclear Reactor Control

2025-12-29

AI Summary

General-purpose AI models exhibit structural deficiencies in physical system control: their perception-centric architectures merely capture input-output statistical correlations, failing to ensure physical consistency of execution outcomes—resulting in unreliable control (e.g., quantitative task accuracy of only 50–53% in nuclear reactor scenarios). This paper proposes a “physics-verification-driven” paradigm for lightweight, agent-based Physical AI, shifting from perceptual imitation to guaranteed physical constraints on action outcome spaces. We identify a variance-collapse phase transition induced by scale expansion, enabling control stabilization. Leveraging a 360M-parameter model, synthetic reactor simulation data (scaling from 10³ to 10⁵ trajectories), physics-constrained policy optimization, and cross-modal representation transfer, we achieve over 500× reduction in control variance; 95% of executions converge to a single robust strategy; 70% of training-distribution behaviors are rejected; and zero-shot transfer across systems and modalities is realized.

Technology Category

• Cognitive Modeling & Cognitive Systems: Agent Architectures
• Multiagent Systems: Adversarial Agents
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

The prevailing paradigm in AI for physical systems, scaling general-purpose foundation models toward universal multimodal reasoning, confronts a fundamental barrier at the control interface. Recent benchmarks show that even frontier vision-language models achieve only 50-53% accuracy on basic quantitative physics tasks, behaving as approximate guessers that preserve semantic plausibility while violating physical constraints. This input unfaithfulness is not a scaling deficiency but a structural limitation. Perception-centric architectures optimize parameter-space imitation, whereas safety-critical control demands outcome-space guarantees over executed actions. Here, we present a fundamentally different pathway toward domain-specific foundation models by introducing compact language models operating as Agentic Physical AI, in which policy optimization is driven by physics-based validation rather than perceptual inference. We train a 360-million-parameter model on synthetic reactor control scenarios, scaling the dataset from 10^3 to 10^5 examples. This induces a sharp phase transition absent in general-purpose models. Small-scale systems exhibit high-variance imitation with catastrophic tail risk, while large-scale models undergo variance collapse exceeding 500x reduction, stabilizing execution-level behavior. Despite balanced exposure to four actuation families, the model autonomously rejects approximately 70% of the training distribution and concentrates 95% of runtime execution on a single-bank strategy. Learned representations transfer across distinct physics and continuous input modalities without architectural modification.

Problem

• Develops domain-specific AI for nuclear reactor control with physics-based validation
• Addresses failure of general AI models in safety-critical physical system control
• Enables stable, reliable control strategies through large-scale synthetic training data

Innovation

• Agentic Physical AI with physics-based validation
• Compact language model for domain-specific foundation
• Autonomous strategy concentration via variance collapse