Research map / Reinforcement learning / Visual and representation RL
What Matters in RL-Based Methods for Object-Goal Navigation? An Empirical Study and A Unified Framework
Hongze Wang, Boyang Sun, Jiaxu Xing, Fan Yang, Marco Hutter, Dhruv Shah, Davide Scaramuzza, Marc Pollefeys
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 Gibson-finetuned Mask R-CNN instead of default checkpoint (ablation) | Corner goal policy · Gibson val | SR (%): 77.7 -> 83.8 | Improved |
| use fine-tuned Mask R-CNN instead of default (ablation) | Frontier policy · Gibson val | SR (%): 71.2 -> 79.9 | Improved |
| use newer detectors (YOLO11-XL, RF-DETR) instead of default Mask R-CNN (ablation) | Corner goal policy · Gibson val | SR (%): 77.7 -> 76.0 (YOLO11-XL), 77.6 (RF-DETR-Seg); SPL 40.9 -> 44.2 | Mixed |
| enlarge local map from 240 to 480 (ablation) | Corner goal policy · Gibson val | SR (%): 78.7 -> 83.8 | Improved |
| enlarge local map from 240 to 480 (ablation) | Frontier policy · Gibson val | SR (%): 79.6 -> 79.9; D-SR 61.0 -> 57.2 | No clear effect |
| add map augmentation (ablation) | Corner goal policy · Gibson val | D-SR (%): 63.5 -> 69.5 | Improved |
| add map augmentation (ablation) | Corner goal policy · Gibson val | SR (%): 83.3 -> 83.8 | No clear effect |
| compress 20-channel semantic map into 3-channel RGB observation (ablation) | RL discrete policy · Gibson val | SR (%): 83.4 -> 83.8 | No clear effect |
| use continuous instead of discrete action space (ablation) | RL policy · Gibson val | SPL (%): 43.3 -> 48.2 | Improved |
| replace CNN with Transformer backbone (ablation) | RL discrete policy · Gibson val | SR (%): 83.7 -> 83.8 | No clear effect |
| use reward 1 (distance to target) instead of reward 2 (ablation) | RL discrete policy · Gibson val | SPL (%): 43.3 -> 45.8; SR 83.8 -> 85.2 | Improved |
| add untrapping helper (ablation) | RL continuous policy · Gibson val | SR (%): 81.3 -> 84.4 | Improved |
Closest papers by meaning
- Active Perception Meets Rule-Guided RL: A Two-Phase Approach for Precise Object Navigation in Complex Environments (ICCV 2025 · similarity 0.90)
- Auxiliary Tasks and Exploration Enable ObjectGoal Navigation (ICCV 2021 · similarity 0.90)
- CogNav: Cognitive Process Modeling for Object Goal Navigation with LLMs (ICCV 2025 · similarity 0.89)
- Trajectory Diffusion for ObjectGoal Navigation (NeurIPS 2024 · similarity 0.87)
- Reasoning in Visual Navigation of End-to-end Trained Agents: A Dynamical Systems Approach (CVPR 2025 · similarity 0.86)
- BeliefMapNav: 3D Voxel-Based Belief Map for Zero-Shot Object Navigation (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.
