A crucial step towards autonomous industrial operation is the ability to create and reconfigure control policies from natural-language requirement specifications with minimal manual redesign. In this context, policy generation by AI agents, when paired with a plant-aware validator (e.g., a digital twin), can be a credible path for checking 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 utilize 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 (ranging from 86.3% to 100% across cases) at a mean inference latency of 3.84 seconds. 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.
Blogger's Review: This study highlights the potential of small language models in industrial control applications, especially in edge computing environments. By integrating validators and self-correction mechanisms, it significantly improves action alignment accuracy, which is of great practical importance. Future research could further explore the integration of different model architectures and algorithms to optimize real-time performance.