Research map / Reinforcement learning / Visual and representation RL
Symmetry-aware Neural Architecture for Embodied Visual Navigation
Shuang Liu, Masanori Suganuma, Takayuki Okatani
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 |
|---|---|---|---|
| replace global policy with symmetry-aware network (G-convolution, blur pooling, SGPP) | ANS · MP3D (trained on Gibson) | area coverage (m^2): 76.3 -> 84.4 | Improved |
| replace global policy with symmetry-aware network | ANS · Gibson (trained on Gibson) | area coverage (m^2): 32.9 -> 33.7 | Improved |
| replace global policy with symmetry-aware network | ANS · MP3D (trained on MP3D) | area coverage (m^2): 80.8 -> 86.2 | Improved |
| replace global policy with symmetry-aware network | ANS · Gibson (trained on MP3D) | area coverage (m^2) | No clear effect |
| learn RL policy over frontiers (FBE-RL) instead of heuristic FBE | FBE · Gibson | area coverage (m^2): 26.5 -> 28.2 | Improved |
| learn RL policy over frontiers (FBE-RL) instead of heuristic FBE | FBE · MP3D (trained on Gibson) | area coverage (m^2): 69.6 -> 63.0 | Hurt |
| add SGPP critic on top of G-conv and blur pooling (S-ANS vs E-ANS) (ablation) | S-ANS · Gibson and MP3D (trained on Gibson) | area coverage (m^2): ranking S-ANS > E-ANS | Improved |
| use global average pooling (G-ANS) instead of no pooling invariance (E-ANS) in critic (ablation) | E-ANS · Gibson and MP3D (trained on Gibson) | area coverage (m^2): ranking E-ANS > G-ANS | Hurt |
| add G-convolution and blur pooling to ANS (G-ANS vs ANS) (ablation) | ANS · Gibson and MP3D (trained on Gibson) | area coverage (m^2): ranking G-ANS > ANS | Improved |
| add symmetry-aware design to critic | ANS · Gibson (critic rotation std) | std of critic output (lower better): 0.115 -> 0.078 | Improved |
Closest papers by meaning
- Symmetry-Aware Neural Architecture for Embodied Visual Exploration (CVPR 2022 · similarity 0.99)
- Learning To Explore Using Active Neural SLAM (ICLR 2020 · similarity 0.88)
- Image2Sim: Scaling Embodied Navigation via Generative Neural Simulator (NeurIPS 2026 · similarity 0.88)
- SiT: Symmetry-invariant Transformers for Generalisation in Reinforcement Learning (ICML 2024 · similarity 0.87)
- Covariance Volume Maximization for Embodied Latent Exploration in Deep Reinforcement Learning (ICML 2026 · similarity 0.87)
- Seeing through Uncertainty: Robust Task-Oriented Optimization in Visual Navigation (NeurIPS 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.
