Closed-Loop Control with Rule-Aligned Small Language Models and Multi-Agent Self-Correction
Quick Answer
This study demonstrates that a compact Small Language Model (Qwen2.5-1.5B) can be effectively retrained for control reasoning in autonomous industrial operations, achieving 91.5% action-alignment accuracy with a mean inference latency of 3.84 seconds.
Quick Take
This study demonstrates that a compact Small Language Model (Qwen2.5-1.5B) can be effectively retrained for control reasoning in autonomous industrial operations, achieving 91.5% action-alignment accuracy with a mean inference latency of 3.84 seconds. The framework integrates a validator-guided correction loop, supporting robust physical regulation in edge applications despite reduced token-level agreement.
Key Points
- Achieved 91.5% average action-alignment accuracy in thermal-control simulations.
- Mean inference latency of 3.84 seconds, suitable for edge applications.
- Framework includes a validator-guided correction loop with a symbolic validation layer.
- Maintained a 95% in-range rate under symbolic re-mapping.
- Supports reconfigurable autonomous control in industrial settings.
Paper Resources
📖 Reader Mode
~2 min readAbstract:A key step toward autonomous industrial operation is the ability to create and reconfigure control policies from natural-language requirement specifications, with minimal or no manual redesign. In this setting, policy generation by AI agents can be a credible path when paired with a plant-aware validator (e.g., a digital twin) that can check generated candidate actions before execution. However, practical deployment is constrained by inference latency and compute footprint: large cloud-based models are often too slow, opaque, or data-sensitive for edge closed-loop use. This work investigates whether a compact Small Language Model (SLM) can be retrained for control reasoning and embedded in a validator-guided correction loop. We use a Qwen2.5-1.5B model aligned via Group Relative Policy Optimization (GRPO), combined with (i) an action agent, (ii) a symbolic/digital-twin-style validation layer, and (iii) a reprompting agent that iteratively steers outputs toward valid actions. In randomized thermal-control simulations (30 experiments with 500 steps each), the framework achieves 91.5% average action-alignment accuracy (86.3%--100% across cases) at 3.84\,s mean inference latency. Under symbolic re-mapping, it maintains a 95% in-range rate, indicating robust physical regulation despite reduced token-level agreement. These results support SLM+validator architectures as a practical path toward reconfigurable autonomous control at the edge.
| Comments: | Accepted by IEEE CCTA 2026 |
| Subjects: | Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA); Robotics (cs.RO) |
| Cite as: | arXiv:2607.09713 [cs.AI] |
| (or arXiv:2607.09713v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2607.09713 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Yuchen Wang [view email]
[v1]
Wed, 24 Jun 2026 13:49:01 UTC (1,259 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.AI
See more →Adversarial Social Epistemology for Assemblies of Humans and Large Language Models
The paper introduces Adversarial Social Epistemology (ASE) to analyze how agents manipulate trust in public communications, highlighting mechanisms that undermine the reliability of testimony and inference. It critiques existing frameworks like epistemic bubbles and misinformation diffusion, proposing a new language for understanding trust breaches and auditing inferential chains in densely interactive environments involving humans and large language models.