Crucible: Quantifying the Potential of Control Algorithms through LLM Agents
This work addresses a gap in control algorithm analysis for domain experts, though it appears incremental as it builds on existing methods with a new evaluation dimension.
The paper tackles the problem of evaluating the tuning potential of control algorithms, which is often overlooked in research focused on ideal configurations, by introducing Crucible, an LLM-driven agent that quantifies this potential across various case studies, leading to performance improvements.
Control algorithms in production environments typically require domain experts to tune their parameters and logic for specific scenarios. However, existing research predominantly focuses on algorithmic performance under ideal or default configurations, overlooking the critical aspect of Tuning Potential. To bridge this gap, we introduce Crucible, an agent that employs an LLM-driven, multi-level expert simulation to turn algorithms and defines a formalized metric to quantitatively evaluate their Tuning Potential. We demonstrate Crucible's effectiveness across a wide spectrum of case studies, from classic control tasks to complex computer systems, and validate its findings in a real-world deployment. Our experimental results reveal that Crucible systematically quantifies the tunable space across different algorithms. Furthermore, Crucible provides a new dimension for algorithm analysis and design, which ultimately leads to performance improvements. Our code is available at https://github.com/thu-media/Crucible.