Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Bayesian Relational Memory for Semantic Visual Navigation
Yi Wu, Yuxin Wu, Aviv Tamar, Stuart Russell, Georgia Gkioxari, Yuandong Tian
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add learned prior over room relations (vs uninformative prior) (ablation) | BRM · House3D RoomNav H=300 | success rate / SPL: 20.9 / 39.4 -> 23.1 / 45.3 | Improved |
| add learned prior over room relations (vs uninformative prior) (ablation) | BRM · House3D RoomNav H=1000 | success rate / SPL: 40.4 / 56.6 -> 41.1 / 57.5 | Improved |
| replace learned LSTM locomotion with oracle locomotion (ablation) | BRM · House3D RoomNav H=1000 | success rate: 41.1 -> 88.6 | Improved |
| replace BRM planner with optimal planner (ablation) | BRM with LSTM locomotion · House3D RoomNav H=1000 | success rate / SPL: 41.1 / 57.5 -> 46.3 / 62.5 | Improved |
| increase memory update interval N from 10 to 30 (ablation) | BRM · House3D RoomNav H=1000 | success rate / SPL: 41.1 / 57.5 -> 29.7 / 35.2 | Hurt |
| increase memory update interval N from 10 to 50 (ablation) | BRM · House3D RoomNav H=1000 | success rate / SPL: 41.1 / 57.5 -> 27.4 / 32.2 | Hurt |
| use ground-truth semantic signals instead of CNN detector (ablation) | BRM · House3D RoomNav H=1000 | success rate / SPL: 41.1 / 57.5 -> 41.7 / 59.8 | No clear effect |
| add BRM planning on top of RNN controller | RNN controller · House3D RoomNav H=300 | success rate / SPL: 19.9 / 34.2 -> 23.1 / 45.3 | Improved |
| add BRM planning on top of RNN controller | RNN controller · House3D RoomNav H=1000 | success rate / SPL: 37.0 / 45.2 -> 41.1 / 57.5 | Improved |
| add BRM on top of RNN controller with terminate action | RNN controller · House3D RoomNav terminate action H=1000 | success rate / SPL: 14.5 / 16.3 -> 17.3 / 23.0 | Improved |
| BRM vs RNN control at planning distance 4, short horizon | RNN controller · House3D RoomNav H=300, plan-steps 4 | success rate / SPL: 5.2 / 15.2 -> 5.8 / 18.6 | Mixed |
Closest papers by meaning
- Hierarchical 3D Scene Graph Construction and Belief-based Planning for Semantic Navigation (ECCV 2026 · similarity 0.89)
- MemoNav: Working Memory Model for Visual Navigation (CVPR 2024 · similarity 0.88)
- Reasoning in Visual Navigation of End-to-end Trained Agents: A Dynamical Systems Approach (CVPR 2025 · similarity 0.88)
- Hierarchical Semantic-Augmented Navigation: Optimal Transport and Graph-Driven Reasoning for Vision-Language Navigation (NeurIPS 2025 · similarity 0.88)
- Planning from Imagination: Episodic Simulation and Episodic Memory for Vision-and-Language Navigation (AAAI 2025 · similarity 0.88)
- 3D-Mem: 3D Scene Memory for Embodied Exploration and Reasoning (CVPR 2025 · similarity 0.87)
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.
