Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Active Neural Localization
Devendra Singh Chaplot, Emilio Parisotto, Ruslan Salakhutdinov
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 |
|---|---|---|---|
| Use learned active actions and a jointly trained neural localization model instead of passive Markov Localization | Maze2D · Maze2D All | accuracy: 0.483 -> 0.934 | Improved |
| Use learned active actions and a jointly trained neural localization model instead of Active Markov Localization (Slow) | Maze2D · Maze2D All | accuracy: 0.904 -> 0.934 | Improved |
| Use learned active actions and a jointly trained neural localization model instead of Active Markov Localization (Fast) | Maze2D · Maze2D All | accuracy: 0.571 -> 0.934 | Improved |
| Replace active Markov search with Active Neural Localization | Maze2D · Maze2D All | runtime for 1000 episodes: 130.2 -> 62.0; 6527.8 -> 62.0 | Mixed |
| Use learned active actions and jointly learned perception instead of Markov Localization | 3D localization · Maze3D and Unreal3D All | accuracy: 0.600 -> 0.826 | Improved |
| Use Active Neural Localization instead of Active Markov Localization (Slow) | 3D localization · Maze3D and Unreal3D All | accuracy: 0.665 -> 0.826 | Improved |
| Use Active Neural Localization instead of Active Markov Localization (Fast) | 3D localization · Maze3D and Unreal3D All | accuracy: 0.632 -> 0.826 | Improved |
| Use Active Neural Localization instead of the 3D Markov baseline | 3D localization · Maze3D and Unreal3D All | runtime for 1000 episodes: 2484.4 -> 905.9 | Improved |
| Train on random-texture Maze3D and transfer directly to Unreal3D without fine-tuning | Active Neural Localization · Maze3D to Unreal3D | accuracy: 0.921 | Improved |
| Test the RGB-based model under changed lighting, using memory images captured with lights on | Active Neural Localization · Unreal3D without lights | accuracy: 0.934 -> 0.505 | Hurt |
| Evaluate on unseen maze layouts and textures, including the no-landmark setting | Active Neural Localization · Maze3D unseen mazes, unseen textures, 0 landmarks | accuracy: 0.871 | Improved |
| Give the agent a longer episode to localize | Active Neural Localization · Maze2D 7x7 | accuracy: 0.936 -> 0.939 | Improved |
Closest papers by meaning
- Learning to Navigate Efficiently and Precisely in Real Environments (CVPR 2024 · similarity 0.85)
- Reasoning in Visual Navigation of End-to-end Trained Agents: A Dynamical Systems Approach (CVPR 2025 · similarity 0.84)
- Active Test-time Vision-Language Navigation (NeurIPS 2025 · similarity 0.84)
- ACTIVE-o3 : Empowering MLLMs with Active Perception via Pure Reinforcement Learning (ICML 2026 · similarity 0.84)
- GOMAA-Geo: GOal Modality Agnostic Active Geo-localization (NeurIPS 2024 · similarity 0.84)
- Seeing through Uncertainty: Robust Task-Oriented Optimization in Visual Navigation (NeurIPS 2025 · similarity 0.83)
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.
