SGA-MCTS: Decoupling Planning from Execution via Training-Free Atomic Experience Retrieval

2026-04-16

AI Summary

Large language models struggle with high inference latency and poor generalization from fine-tuning in complex, multi-step decision-making tasks. This work proposes a training-free, non-parametric planning approach that decouples planning from execution: offline, it leverages Monte Carlo Tree Search to generate delexicalized state-goal-action (SGA) atoms; online, it dynamically re-ground these atoms via a hybrid symbolic-semantic retrieval mechanism to construct soft reasoning prompts. Operating with frozen model weights, the method effectively balances the depth of System 2 reasoning with the speed of System 1 inference. Evaluated across multiple challenging benchmarks, the approach enables open-source, frozen models to match the performance of advanced systems such as GPT-5, all without task-specific fine-tuning.

Technology Category

• Planning, Routing, and Scheduling: Planning with Language Models
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Search and Retrieval-Augmented AI: Large language models for search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

LLM-powered systems require complex multi-step decision-making abilities to solve real-world tasks, yet current planning approaches face a trade-off between the high latency of inference-time search and the limited generalization of supervised fine-tuning. To address this limitation, we introduce \textbf{SGA-MCTS}, a framework that casts LLM planning as non-parametric retrieval. Offline, we leverage Monte Carlo Tree Search (MCTS) to explore the solution space and distill high-fidelity trajectories into State-Goal-Action (SGA) atoms. These atoms are de-lexicalized primitives that abstract concrete entities into symbolic slots, preserving reusable causal logic while discarding domain-specific noise. Online, a retrieval-augmented agent employs a hybrid symbolic-semantic mechanism to fetch relevant SGAs and re-ground them into the current context as soft reasoning hints. Empirical results on complex benchmarks demonstrate that this paradigm enables frozen, open-weights models to match the performance of SOTA systems (e.g., GPT-5) without task-specific fine-tuning. By effectively amortizing the heavy computational cost of search, SGA-MCTS achieves System 2 reasoning depth at System 1 inference speeds, rendering autonomous planning both scalable and real-time feasible.

Problem

• LLM planning
• multi-step decision-making
• inference latency
• generalization
• real-time planning

Innovation

• SGA-MCTS
• non-parametric retrieval
• atomic experience
• symbolic-semantic reasoning
• amortized planning