Constructing coherent spatial memory in LLM agents through graph rectification

Given a map description through global traversal navigation instructions, an LLM can often infer the implicit spatial layout and answer user queries by providing shortest paths.However, such context-dependent querying becomes incapable as environments grow larger, motivating the need for incremental map construction that builds a complete topological graph from stepwise observations.We propose a framework for LLM-driven construction and map repair, designed to detect, localize, and correct structural inconsistencies in incrementally constructed navigation graphs.Our contributions include a version control mechanism for graph construction, an Edge Impact Score for repair prioritization, and a cleaned variant of the MANGO benchmark tailored for LLMdriven map construction and repair.Compared with direct LLM-based incremental mapping, MapRepair raises node recall by 8.6 percentage points to 94.3% and edge recall by 55.8 percentage points to 88.2%, evaluated on Chapters 16 and 17 of Dream of the Red Chamber with GPT-4.1 as the underlying LLM.