Multimodal Agent MemorySubmit a paper

EmbodiedLGR-Agent

EmbodiedLGR: Integrating Lightweight Graph Representation and Retrieval for Semantic-Spatial Memory in Robotic Agents

Paolo Riva, Leonardo Gargani, Matteo Frosi, Matteo Matteucci · 2026-04-20

Read the paperPDF

Management
T2 Dynamic Memory
Memory changes during deployment through a stable management policy.
Primary modality
Embodied & 3D
Also: text
Representation
Cross-represented explicit

How experience becomes memory

Each path is one memory component, traced from what is stored to how it is found again.

  1. semantic object-position graph

    1. Represented asSpatial, temporal & cross-modal structured memoryCross-represented explicit
    2. Stored inGraph / relational store
    3. Retrieved byStructure traversal
    4. Source evidenceSource linked & recoverable
    Low-level objects and positions are stored in a semantic graph built from robot observations.Full paper: III. METHOD · PDF p. 2 · Figure 1

  2. scene-description memory

    1. Represented asPerception-to-text records & summariesCross-represented explicit
    2. Stored inVector-indexed collection
    3. Retrieved bySimilarity search
    4. Source evidenceSource linked & recoverable
    High-level observed-scene descriptions are retained in a RAG store alongside the graph.Full paper: III. METHOD · PDF p. 2 · Figure 1

How memory changes (T2 operations)

  • Update · Rule-based update

    The fixed local building pipeline adds observed objects, positions, and scene descriptions to memory. For EmbodiedLGR-Agent, this specifically supports fixed-rule insertion or replacement of semantic object-position graph.

  • Reorganization · Relinking & reindexing

    New observations are linked into the semantic graph and retrieval store. For EmbodiedLGR-Agent, this specifically supports relinking or reindexing of stored items of semantic object-position graph.

Related systems