Multimodal Agent MemorySubmit a paper

DGSG-Mind

DGSG-Mind: Dynamic 3D Gaussian Scene Graphs for Long-Term Scene Understanding and Grounding

Luzhou Ge, Xiangyu Zhu, Jinyan Liu, Xuesong Li · 2026-05-29

Read the paperPDF

Management
T2 Dynamic Memory
Memory changes during deployment through a stable management policy.
Primary modality
Embodied & 3D
Also: text, Image, video
Representation
Hybrid
Distinct operative representation paths are deliberately linked or jointly consumed by downstream reasoning, generation, or control.

How experience becomes memory

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

  1. long-term scene representation: object instances

    1. Represented asModality-specific encodings & geometric stateSource- and modality-specific
    2. Stored inGraph / relational store
    3. Retrieved bySimilarity search, Structure traversal
    4. Source evidenceRetained but not accessible to the agent
    Dynamic 3D Gaussian Splatting representation combined with scene graph nodes and edges. language-grounded retrieval over the scene graph for grounding and long-term understanding.Full-paper method / memory representation and retrieval · PDF p. 2 · Figure 1 / method overview

  2. spatial relationships

    1. Represented asSpatial, temporal & cross-modal structured memoryCross-represented explicit
    2. Stored inGraph / relational store
    3. Retrieved bySimilarity search, Structure traversal
    4. Source evidenceSource not retained
    Dynamic 3D Gaussian Splatting representation combined with scene graph nodes and edges. language-grounded retrieval over the scene graph for grounding and long-term understanding.Full-paper method / memory representation and retrieval · PDF p. 2 · Figure 1 / method overview

How memory changes (T2 operations)

  • Update · Rule-based update

    During operation, DGSG-Mind applies fixed-rule insertion or replacement to long-term scene representation: object instances; the memory state changes while the governing update policy remains fixed.

  • Revision · Rule-based state correction

    During operation, DGSG-Mind applies fixed-rule correction of stored state to long-term scene representation: object instances; the memory state changes while the governing update policy remains fixed.

  • Forgetting · Discrete eviction

    Online updates as the environment evolves; long-term adaptation. For DGSG-Mind, this specifically supports explicit deletion or eviction of stored items of long-term scene representation: object instances.

  • Reorganization · Relinking & reindexing

    Online updates as the environment evolves; long-term adaptation. For DGSG-Mind, this specifically supports relinking or reindexing of stored items of long-term scene representation: object instances.

Related systems