Reproduce “SDGym: Low-Code Reinforcement Learning Environments using System Dynamics Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Emmanuel Klu, Sameer Sethi, DJ Passey, Donald Martin Jr
SDGym is a low-code Python library designed to transform System Dynamics models into Reinforcement Learning environments, aiming to improve the realism and ethical considerations of AI. The framework demonstrated its capability to generate complex environments from existing SD models and enabled an RL agent to learn dynamic intervention policies for a real-world system, such as electric vehicle adoption.
Key insights
- System Dynamics models, with their emphasis on feedback loops and non-linear interactions, can provide a robust, causally explicit foundation for creating realistic and complex Reinforcement Learning environments.
- Reinforcement Learning agents can effectively learn dynamic, adaptive intervention strategies within SD models, moving beyond static policy analysis to discover optimal, time-varying policies.
- Bridging Reinforcement Learning and System Dynamics through a tool like SDGym accelerates interdisciplinary research, enabling more robust, fair, and transparent AI system development.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.