LLM Planning

LLM: Large Language Model

Overview

Large language model planning asks whether models can turn goals and instructions into ordered actions, then adapt those actions as constraints or environments change. Common approaches include chain-of-thought and tree search, hierarchical task decomposition, reinforcement learning, and hybrid systems that connect language models to PDDL, symbolic solvers, tools, or learned world models. Benchmarks such as BlocksWorld, TravelPlanner, and ALFWorld test plan validity, efficiency, and adaptation in settings from fixed puzzles to interactive environments. Models still struggle with long horizons, coupled constraints, and disruptions; verification and solver-based methods improve reliability but raise questions about cost, generalization, and how much planning models can do on their own.

Latest papers

151 papers11 papers in the last four weeks, down 58% on the four weeks before. 0.1% of all new papers.

Jan 30, 2025

Successor-Generator Planning with LLM-generated Heuristics

Heuristics are a central component of deterministic planning, particularly in domain-independent settings where general applicability is prioritized over task-specific tuning. This work revisits that paradigm in light of recent advances in large language models (LLMs), which enable the automatic synthesis of heuristics directly from problem definitions -- bypassing the need for handcrafted domain knowledge. We present a method that employs LLMs to generate problem-specific heuristic functions from planning tasks specified through successor generators, goal tests, and initial states written in a general-purpose programming language. These heuristics are compiled and integrated into standard heuristic search algorithms, such as greedy best-first search. Our approach achieves competitive, and in many cases state-of-the-art, performance across a broad range of established planning benchmarks. Moreover, it enables the solution of problems that are difficult to express in traditional formalisms, including those with complex numeric constraints or custom transition dynamics. We provide an extensive empirical evaluation that characterizes the strengths and limitations of the approach across diverse planning settings, demonstrating its effectiveness.