Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Emergence of Maps in the Memories of Blind Navigation Agents
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Navigate using only egomotion sensing and recurrent memory | Blind LSTM agent · PointGoalNav, unseen Matterport3D test scenes | Success: 95.1 ±1.3 | Helped |
| Use recurrent memory rather than a one-step memory budget (k=1) (ablation) | Blind LSTM agent · PointGoalNav | Success: nearly 0% at k=1 | Helped |
| Increase memory budget (ablation) | Blind LSTM agent · PointGoalNav | Navigation performance: one thousand steps | Helped |
| Initialize the probe with trained agent episodic memory instead of no informative memory (ablation) | Blind LSTM probe · Probe S-to-T, unseen environments | SPL: 62.9% ±1.6% -> 85.0% ±1.6% | Helped |
| Use trained agent memory rather than untrained-agent memory or all-zero memory (ablation) | Blind LSTM probe · Probe navigation | SPL: +13 points SPL | Helped |
| Use trained agent memory rather than untrained-agent memory or all-zero memory (ablation) | Blind LSTM probe · Probe navigation | Success: +4 points Success | Helped |
| Replace trained memory with untrained-agent memory (ablation) | Blind LSTM probe · Probe navigation | SPL | No clear effect |
| Decode occupancy maps from trained rather than untrained agent memory (ablation) | Occupancy map decoder · Novel validation scenes | IoU: 12.5% vs 32.5% | Helped |
| Use a minimal blind sensor suite rather than depth sensing (ablation) | Navigation agent · PointGoalNav | Success: 95.1 vs 94.0 | No clear effect |
| Use a minimal blind sensor suite rather than depth sensing (ablation) | Navigation agent · PointGoalNav | SPL: 62.9 vs 83.0 | Hurt |
| Train in one environment with consistent global coordinates rather than many environments with episodic coordinates (ablation) | Blind LSTM agent · Navigation in novel test environments | Success: 100% on train and 0% on test | Hurt |
| Train in one scene with global coordinates (ablation) | Blind LSTM agent · Single-scene training | SPL: 99 ±0.1 | Helped |
Closest papers by meaning
- Reasoning in Visual Navigation of End-to-end Trained Agents: A Dynamical Systems Approach (CVPR 2025 · similarity 0.89)
- Planning from Imagination: Episodic Simulation and Episodic Memory for Vision-and-Language Navigation (AAAI 2025 · similarity 0.87)
- Learning with a Mole: Transferable latent spatial representations for navigation without reconstruction (ICLR 2024 · similarity 0.87)
- Time Makes Space: Emergence of Place Fields in Networks Encoding Temporally Continuous Sensory Experiences (NeurIPS 2024 · similarity 0.86)
- Active Exploring like a Pigeon: Reinforcing Spatial Reasoning via Agentic Vision-Language Models (ICML 2026 · similarity 0.86)
- Intrinsic Goals for Autonomous Agents: Model-Based Exploration in Virtual Zebrafish Predicts Ethological Behavior and Whole-Brain Dynamics (NeurIPS 2025 · similarity 0.86)
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.