Multimodal Spatial Language Maps
Multimodal spatial language maps for robot navigation and manipulation
Chenguang Huang, Oier Mees, Andy Zeng, Wolfram Burgard · 2025-01-01
- Management
- T1 Static Memory
- Memory is built once and reused without change during deployment.
- 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.
semantic map / spatial language map
- Represented asModality-specific encodings & geometric stateSource- and modality-specific
- Stored inFlat sequential records
- Retrieved byMetadata / spatiotemporal retrieval, Structure traversal
- Source evidenceRetained but not accessible to the agent
Autonomous exploration fuses foundation-model features with 3D reconstruction. LLMs translate commands into spatial goals localized in the map.Full-paper method / memory representation and retrieval · PDF p. 2 · Figure 1 / method overview
semantic map / spatial language map
- Represented asExperience-conditioned parametric memoryInternalized latent & parametric
- Stored inModular adapter bank
- Retrieved byMetadata / spatiotemporal retrieval, Structure traversal, Latent / parametric activation
- Source evidenceSource not retained
Autonomous exploration fuses foundation-model features with 3D reconstruction. LLMs translate commands into spatial goals localized in the map.Full-paper method / memory representation and retrieval · PDF p. 2 · Figure 1 / method overview
Related systems
- Affordance RAGT1 · Embodied & 3D
- DRAET1 · Embodied & 3D
- HIMMT2 · Embodied & 3D
- StoryAgentT1 · Video & streaming
- Vision to Geometry / 3DSPMRT2 · Embodied & 3D
- C-NavT1 · Embodied & 3D