Research map / Label-efficient and robust learning / Self-supervised and contrastive
Learning to Learn How to Learn: Self-Adaptive Visual Navigation Using Meta-Learning
Check 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 |
|---|---|---|---|
| adapt the policy during navigation using a learned self-supervised interaction loss | A3C visual navigation · AI2-THOR novel scenes, All | Success: 33.04 -> 40.86 | Helped |
| adapt the policy during navigation using a learned self-supervised interaction loss | A3C visual navigation · AI2-THOR novel scenes, All | SPL: 14.68 -> 16.15 | Helped |
| replace the learned interaction loss with a hand-crafted prediction loss (ablation) | SAVN · AI2-THOR novel scenes, All | SPL: 16.15 -> 14.36 | Hurt |
| replace the learned interaction loss with a hand-crafted prediction loss (ablation) | SAVN · AI2-THOR novel scenes, All | Success: 40.86 -> 38.06 | Hurt |
| replace the learned interaction loss with a hand-crafted diversity loss (ablation) | SAVN · AI2-THOR novel scenes, All | SPL: 16.15 -> 15.12 | Hurt |
| replace the learned interaction loss with a hand-crafted diversity loss (ablation) | SAVN · AI2-THOR novel scenes, All | Success: 40.86 -> 39.52 | Hurt |
| add self-attention memory over the latest k=6 LSTM hidden states (ablation) | No Adapt (A3C) · AI2-THOR novel scenes, All | Success: 33.04 -> 34.30 | Helped |
| add self-attention memory over the latest k=6 LSTM hidden states (ablation) | No Adapt (A3C) · AI2-THOR novel scenes, L ≥ 5 | Success: 21.44 -> 20.50 | Hurt |
| add the hand-crafted prediction loss to A3C training (ablation) | No Adapt (A3C) · AI2-THOR novel scenes, All | Success: 33.04 -> 34.80 | Helped |
| add the hand-crafted prediction loss to A3C training (ablation) | No Adapt (A3C) · AI2-THOR novel scenes, All | SPL: 14.68 -> 14.95 | No clear effect |
| perform more interaction-gradient updates during training and testing (ablation) | SAVN · AI2-THOR navigation | Success rate | Helped |
| use a schedule for the interaction-gradient interval k (ablation) | SAVN · AI2-THOR navigation | Performance | No clear effect |
Closest papers by meaning
- Unsupervised Reinforcement Learning of Transferable Meta-Skills for Embodied Navigation (CVPR 2020 · similarity 0.90)
- Unsupervised Curricula for Visual Meta-Reinforcement Learning (NeurIPS 2019 · similarity 0.88)
- Fast And Slow Learning Of Recurrent Independent Mechanisms (ICLR 2021 · similarity 0.88)
- Test-Time Adaptation for Online Vision-Language Navigation with Feedback-based Reinforcement Learning (ICML 2025 · similarity 0.87)
- C-NAV: Towards Self-Evolving Continual Object Navigation in Open World (NeurIPS 2025 · similarity 0.86)
- Reasoning in Visual Navigation of End-to-end Trained Agents: A Dynamical Systems Approach (CVPR 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.