Abstract
We study the problem of generating plans for given natural language planningtask requests. On one hand, LLMs excel at natural language processing but donot perform well on planning. On the other hand, classical planning tools excelat planning tasks but require input in a structured language such as thePlanning Domain Definition Language (PDDL). We leverage the strengths of boththe techniques by using an LLM for generating the PDDL representation (taskPDDL) of planning task requests followed by using a classical planner forcomputing a plan. Unlike previous approaches that use LLMs for generating taskPDDLs directly, our approach comprises of (a) translate: using an LLM only forgenerating a logically interpretable intermediate representation of naturallanguage task descriptions, (b) infer: deriving additional logically dependentinformation from the intermediate representation using a logic reasoner(currently, Answer Set Programming solver), and (c) compile: generating thetarget task PDDL from the base and inferred information. We observe that usingan LLM to only output the intermediate representation significantly reduces LLMerrors. Consequently, TIC approach achieves, for at least one LLM, highaccuracy on task PDDL generation for all seven domains of our evaluationdataset.