Research map / Reinforcement learning / Visual and representation RL
Active Perception Meets Rule-Guided RL: A Two-Phase Approach for Precise Object Navigation in Complex Environments
Liang Qin, Min Wang, Peiwei Li, Wengang Zhou, Houqiang Li
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 active last-mile navigation module to the rule-based agent (ablation) | Rule* · OVMM ObjectNav (YOLO-World+SAM2 perception) | SR (%): 51.1 -> 60.2 | Improved |
| add active last-mile navigation module to the RL agent (ablation) | RL* · OVMM ObjectNav (YOLO-World+SAM2 perception) | SR (%): 40.8 -> 43.6 | Improved |
| add active last-mile navigation module to the rule-based agent (ablation) | Rule* · OVMM ObjectNav (GT semantic masks) | SR (%): 54.4 -> 69.3 | Improved |
| add active last-mile navigation module to the RL agent (ablation) | RL* · OVMM ObjectNav (GT semantic masks) | SR (%): 48.5 -> 56.9 | Improved |
| add active last-mile navigation module to the rule-based agent (ablation) | Rule* · OVMM ObjectNav (GT semantic masks) | PIC: 0.0162 -> 0.0421 | Improved |
| add rule-guided RL exploration over the rule-based exploration with last-mile module (ablation) | Rule*+L-M Nav · OVMM ObjectNav (YOLO-World+SAM2 perception) | SR (%): 60.2 -> 64.5 | Improved |
| add rule-guided RL exploration over RL exploration with last-mile module (ablation) | RL*+L-M Nav · OVMM ObjectNav (GT semantic masks) | SR (%): 56.9 -> 74.2 | Improved |
| lower YOLO-World confidence threshold from 0.5 to 0.3 (ablation) | APRR · OVMM ObjectNav (YOLO-World+SAM2 perception) | SR (%): 64.5 -> 63.3 | Hurt |
| increase last-mile turn angle from 15 to 30 degrees (ablation) | APRR · OVMM ObjectNav (GT semantic masks) | SR (%): 74.2 -> 68.3 | Hurt |
| use ground-truth semantic masks instead of detector | APRR · OVMM ObjectNav | SR (%): 64.5 -> 74.2 | Improved |
Closest papers by meaning
- What Matters in RL-Based Methods for Object-Goal Navigation? An Empirical Study and A Unified Framework (ECCV 2026 · similarity 0.90)
- ACTIVE-o3 : Empowering MLLMs with Active Perception via Pure Reinforcement Learning (ICML 2026 · similarity 0.88)
- ActiveVLA: Injecting Active Perception into Vision-Language-Action Models for Precise 3D Robotic Manipulation (CVPR 2026 · similarity 0.87)
- SPIN: Simultaneous Perception Interaction and Navigation (CVPR 2024 · similarity 0.86)
- Real-World Reinforcement Learning of Active Perception Behaviors (NeurIPS 2025 · similarity 0.86)
- A Single Goal is All You Need: Skills and Exploration Emerge from Contrastive RL without Rewards, Demonstrations, or Subgoals (ICLR 2025 · similarity 0.85)
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.
