Multimodal Agent MemorySubmit a paper

MEMORA

MEMORA: Embodied Action Memory from Egocentric Videos for Reasoning and Planning

Zihao Yu, Xiu Yuan, Chongjie Zhang · 2026-07-15

Read the paperPDF

Management
T2 Dynamic Memory
Memory changes during deployment through a stable management policy.
Primary modality
Embodied & 3D
Also: Video & streaming, 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. environment, entity, activity, and inferred-knowledge stores

    1. Represented asSpatial, temporal & cross-modal structured memoryCross-represented explicit
    2. Stored inGraph / relational store, Hierarchical multi-tier store
    3. Retrieved byMetadata / spatiotemporal retrieval, Structure traversal, Planner / tool-assisted retrieval
    4. Source evidenceSource not retained
    Four typed stores retain object states, activities, procedures, and regularities.Full paper: Memory · PDF p. 4

How memory changes (T2 operations)

  • Update · Prompted LLM/VLM record update

    Online editing maintains object identities and state histories. For MEMORA, this specifically supports LLM-mediated memory editing of environment, entity, activity, and inferred-knowledge stores.

  • Consolidation · Episodic / hierarchical summarization

    Offline consolidation forms reusable procedures. For MEMORA, this specifically supports episodic or hierarchical summarization of environment, entity, activity, and inferred-knowledge stores.

  • Consolidation · Experience abstraction

    Repeated experience is abstracted into procedures and participant-specific regularities. For MEMORA, this specifically supports abstraction of episodes into reusable experience of environment, entity, activity, and inferred-knowledge stores.

Related systems