Research map / 3D vision / 3D generation
Space-Aware World Models: Spatial Persistence Through Factored Scene Representations
Bruno Corcuera, Carlos Eiras-Franco, Brais Cancela
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes space-aware world models that factor scene layout, appearance, and agent motion into separate representations to improve spatial persistence in long-horizon navigation simulation.
Problem
Long-horizon navigation world models must reproduce the same landmarks when physical states are revisited, but single-latent-state latent-dynamics models suffer brittle revisitation consistency under compounding errors.
Approach
Replaces a single predictive latent state that jointly encodes layout, appearance, and agent motion with a structured, factored scene representation so that spatial persistence is maintained explicitly.
Experimental setup
Baselines: standard latent-dynamics world models
Closest papers by meaning
- Drift-Resistant Navigation World Model with Anchored Epipolar Guidance (NeurIPS 2026 · similarity 0.89)
- WorldMem: Long-term Consistent World Simulation with Memory (NeurIPS 2025 · similarity 0.89)
- StateSpaceDiffuser: Bringing Long Context to Diffusion World Models (NeurIPS 2025 · similarity 0.89)
- Learning 3D Persistent Embodied World Models (NeurIPS 2025 · similarity 0.89)
- Video World Models with Long-term Spatial Memory (NeurIPS 2025 · similarity 0.89)
- LoopNav: Benchmarking Spatial Consistency in World Models (NeurIPS 2026 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
