Regelum: Graph Dependency Resolution and Execution Orchestration for Control Systems

The rapid advancement of control theory and reinforcement learning has created a need for flexible, modular frameworks that support complex research pipelines. Current solutions present researchers with a challenging tradeoff: choosing between open-source frameworks with limited modularity or comprehensive proprietary tools with restricted customization. We introduce Regelum, with a focus on its graph dependency resolution and execution orchestration system that efficiently manages interconnected computational components. By implementing a robust algorithm for topological sorting, cycle detection, and time synchronization, Regelum enables researchers to define self-contained mdes within directed graphs while automating the complex task of determining execution order and data flow. Wed emonstrate the system's capabilities through a case study on Lyapunov-based adaptive control, showing how the dependency resolution mechanism facilitates experimentation with advanced control strategies and sim-to-real transitions. The technical approach to graph management in Regelum addresses critical gaps in existing frameworks, offering a path toward more reproducible and customizable research in adaptive control and reinforcement learning.