Using Large Language Models for Abstraction of Planning Domains - Extended Version
This addresses the problem of improving planning and reasoning for AI agents by automating abstraction generation, though it is incremental as it builds on existing LLM and planning methods.
The paper tackles the challenge of generating abstractions for dynamic planning domains by using large language models (LLMs) with in-context learning to produce abstract PDDL domains and problem instances from natural language objectives, and finds that GPT-4o can synthesize useful abstractions in simple settings, performing better on actions than fluents.
Generating an abstraction of a dynamic domain that aligns with a given purpose remains a significant challenge given that the choice of such an abstraction can impact an agent's ability to plan, reason, and provide explanations effectively. We model the agent's concrete behaviors in PDDL and investigate the use of in-context learning with large language models (LLMs) for the generation of abstract PDDL domains and problem instances, given an abstraction objective specified in natural language. The benchmark examples we use are new and have not been part of the data any LLMs have been trained on. We consider three categories of abstractions: abstraction of choice of alternative concrete actions, abstraction of sequences of concrete actions, and abstraction of action/predicate parameters, as well as combinations of these. The generated abstract PDDL domains and problem instances are then checked by symbolic validation tools as well as human experts. Our experiments show that GPT-4o can generally synthesize useful planning domain abstractions in simple settings, although it is better at abstracting over actions than over the associated fluents.