Multimodal Agent MemorySubmit a paper

Skill-3D

Skill-3D: Evolving Scene-Aware Skills for Agentic 3D Spatial Reasoning

Haoyuan Li, Zhengdong Hu, Jun Wang, Hehe Fan, Yi Yang · 2026-06-11

Read the paperPDF

Management
T1 Static Memory
Memory is built once and reused without change during deployment.
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. scene-aware procedural skills

    1. Represented asSpatial, temporal & cross-modal structured memoryCross-represented explicit
    2. Stored inFlat sequential records
    3. Retrieved bySimilarity search, Planner / tool-assisted retrieval
    4. Source evidenceSource not retained
    Scene Memory records rollouts; reusable scene-aware skills are distilled into a skill library. retrieves compact skill subsets by task category, target entities, scene signature, required evidence, historical success, failure lessons, and tool cost.Full paper: Scene Memory · PDF p. 4 · Figure 1 / method overview

Related systems