Planning with means-ends analysis helps AI in continuous action tasks
Means-Ends Analysis Helps LLMs Plan in Continuous-State Domains
Artificial IntelligenceMachine Learning
Summary
Planning with AI usually works best when the possible actions are limited and easy to list. The authors show that this approach struggles when the actions can vary continuously, like mixing liquids. They use an old problem-solving idea called means-ends analysis to help AI break goals into smaller steps and keep track of progress. Their method improves AI planning in a test environment where actions are continuous, showing AI can reason better this way.
What this means in practice
- •For robotics developers: Improve robot planning in tasks requiring continuous control, like liquid handling or precise manipulation, by structuring AI reasoning with means-ends analysis.
- •For industrial process engineers: Design better AI-driven process controls for continuous-state manufacturing steps such as mixing or chemical blending using improved AI planning methods.
Authors
Jiayi Li, Carl Kingsford
Abstract
LLM planning research has traditionally focused on discrete-state domains, where actions form a finite, enumerable set. Existing LLM planning frameworks largely assume actions are enumerable, modeling planning as a search process over candidate actions. However, in continuous-state domains, where actions form a continuum and cannot be enumerated, this assumption breaks down. We find a remedy in Newell and Simon's Means-Ends Analysis (MEA), a general problem-solving technique rooted in heuristics from human psychology. We develop Medaka, an MEA-inspired structured prompting framework that encourages recursive goal decomposition, algebraic derivation, and per-step state tracking. To evaluate native LLM planning abilities in continuous-state domains, we develop Liquidworld, a systematic testbed on liquid mixing. On Liquidworld, Medaka can improve performance across model families, suggesting that classical means-ends analysis elicits LLM-native reasoning abilities that are helpful for continuous-state planning.