Multimodal Agent MemorySubmit a paper

HAM-VLN

HAM-VLN: Harnessing Hierarchical Agentic Memory for Zero-Shot Vision-and-Language Navigation

An Liu, Bingxi Liu, Hongyu Ding, Yixuan Jiang, Yaran Chen, Fulin Tang, et al. · 2026-07-31

Read the paperPDF

Management
T2 Dynamic Memory
Memory changes during deployment through a stable management policy.
Primary modality
Embodied & 3D
Also: text
Representation
Hybrid
Distinct operative representation paths are explicitly linked for downstream retrieval, reasoning, or control.

How experience becomes memory

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

  1. recent waypoint observations

    1. Represented asRetained perceptual source evidenceSource- and modality-specific
    2. Stored inHierarchical multi-tier store
    3. Retrieved byMetadata / spatiotemporal retrieval
    4. Source evidenceSource directly available
    Recent depth-grounded waypoints remain verbatim within a bounded window.Full paper: Memory · PDF p. 3 · Figure 1

  2. persistent world graph and reflection notes

    1. Represented asSpatial, temporal & cross-modal structured memoryCross-represented explicit
    2. Stored inGraph / relational store
    3. Retrieved byStructure traversal, Similarity search
    4. Source evidenceSource linked & recoverable
    The agent records room, object, progress, and failure information in a depth-grounded graph.Full paper: Method · PDF p. 2 · Figure 1

How memory changes (T2 operations)

  • Update · Rule-based update

    Each action decision also writes semantic and reflective information into the world graph. For HAM-VLN, this specifically supports fixed-rule insertion or replacement of recent waypoint observations.

  • Forgetting · Discrete eviction

    Older verbatim waypoints leave the bounded recent window. For HAM-VLN, this specifically supports explicit deletion or eviction of stored items of recent waypoint observations.

  • Reorganization · Relinking & reindexing

    Older history is re-accessed by relevance, recency, salience, and one-hop graph expansion. For HAM-VLN, this specifically supports relinking or reindexing of stored items of recent waypoint observations.

Related systems