MemPlan: Memory-Conditioned PDDL Planning for LLM Agents
October 23, 2026
MemPlan uses LLMs as bridges to symbolic PDDL planning in partially observable environments. It utilizes positive graph memory for plausible hypothesis costs and negative memory to record refuted assumptions as exclusion constraints.
HOW THIS AFFECTS YOU
●
builderYou can implement this framework to reduce hallucinated actions in long-horizon agent workflows.
●
researcherThis approach provides a way to combine the flexibility of LLMs with the rigor of symbolic planning.