Multimodal Agent MemorySubmit a paper

VISOR

VISOR: Visual Evidence Space for Robust GUI and Visual Agent Reasoning

Yucheng Shen, Jiulong Wu, Jizhou Huang, Dawei Yin, Lingyong Yan, Min Cao · 2026-04-10

Read the paperPDF

Management
T2 Dynamic Memory
Memory changes during deployment through a stable management policy.
Primary modality
Image
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 visual trajectory window

    1. Represented asRetained perceptual source evidenceSource- and modality-specific
    2. Stored inFlat sequential records
    3. Retrieved byMetadata / spatiotemporal retrieval
    4. Source evidenceSource directly available
    A dynamic trajectory window preserves recent visual observations for action evaluation.Full paper: 2.1 Visual Retrieval-Augmented Generation · PDF p. 2 · Figure 1

  2. structured Evidence Space

    1. Represented asSpatial, temporal & cross-modal structured memoryCross-represented explicit
    2. Stored inGraph / relational store
    3. Retrieved byStructure traversal, Planner / tool-assisted retrieval
    4. Source evidenceSource linked & recoverable
    The Evidence Space organizes task-relevant observations and intents for grounded correction.Full paper: 3 Method · PDF p. 3 · Figure 1

How memory changes (T2 operations)

  • Update · Rule-based update

    During operation, VISOR applies fixed-rule insertion or replacement to recent visual trajectory window; the memory state changes while the governing update policy remains fixed.

  • Consolidation · Episodic / hierarchical summarization

    During operation, VISOR applies episodic or hierarchical summarization to recent visual trajectory window; the memory state changes while the governing update policy remains fixed.

  • Forgetting · Discrete eviction

    A sliding trajectory window bounds retained visual history. For VISOR, this specifically supports explicit deletion or eviction of stored items of recent visual trajectory window.

Related systems