Reproduce “WOFOSTGym: A Crop Simulator for Learning Annual and Perennial Crop Management Strategies” 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
William Solow, Sandhya Saisubramanian, Alan Fern
Researchers from Oregon State University developed WOFOSTGym, a high-fidelity crop simulation environment built on the WOFOST Crop Growth Model, to train reinforcement learning agents for complex agricultural management. It supports both annual and perennial crops, multi-farm scenarios, and tasks with partial observability and delayed feedback, demonstrated by experiments showing its utility for RL and imitation learning in constrained and unconstrained settings.
Key insights
- Existing off-the-shelf reinforcement learning and imitation learning algorithms face limitations in effectively managing long-horizon perennial crop cycles and adhering to strict agricultural constraints.
- Providing RL agents with access to more relevant observable state features (e.g., daily rainfall, soil surface fertilizer) leads to improved constraint adherence and better outcomes in agromanagement tasks.
- High-fidelity simulators like WOFOSTGym, which incorporate specific modifications for perennial crops and advanced parameter calibration via Bayesian optimization, are crucial for bridging the gap between theoretical RL and practical agricultural applications.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.