Mimir enables safe long-term irrigation control using physics and llm reasoning
Mimir: Physics-Grounded LLM Agents for Long-Horizon Irrigation Control
Artificial Intelligence
Summary
Managing irrigation over an entire growing season is tricky because decisions made one day affect soil and plants later on, and mistakes can add up over time. The authors created Mimir, a computer agent that combines a physics-based model with large language model reasoning to plan irrigation safely and effectively. Mimir tests and revises its plans before acting and learns from past failures without changing the core physical rules. This approach led to less water use and better control than traditional methods in simulations across different locations and crops.
What this means in practice
- •For farm managers: Improve irrigation scheduling over entire growing seasons to reduce water use without compromising crop health.
- •For industrial control engineers: Design autonomous control systems that safely manage long-term physical processes where actions have delayed and compounded effects.
Authors
Yimeng Liu, Mi Zhang, Younsuk Dong, Zhichao Cao
Abstract
Large language model (LLM) agents increasingly combine reasoning, tool use, and action, but most evidence comes from episodic tasks with relatively immediate feedback and reset failures. Long-running physical control operates in a different regime: actions alter future states, errors compound across decisions, and an agent must improve from experience without being allowed to rewrite the physical rules that make execution safe. We study this regime through irrigation, where daily decisions interact with soil-water dynamics over entire growing seasons. We present Mimir, a physics-grounded LLM agent organized around two repair timescales. At the fast timescale, a structured physical interface and deterministic simulator turn an LLM output into a proposal that we numerically check, revise, and subject to bounded deterministic action selection before execution. At the slow timescale, recurrent failure patterns are consolidated into persistent contextual principles that condition future proposals, while the physical model, evaluator, and execution constraints remain immutable. Under a common retrospective evaluator across multiple sites, crops, and years, Mimir attains the lowest reported aggregate control cost among the evaluated references and uses about 51% less irrigation than the historical schedule replay. The ablation study show higher control cost when forward simulation, verified revision, or persistent context is removed; model-scale and model-family studies show no monotonic gain from increasing LLM size. The resulting lesson show that persistent physical agents can combine semantic reasoning with bounded, evidence-driven self-improvement while reserving physical truth and actuator authority for explicit numerical mechanisms.