Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Towards Deviation-Robust Agent Navigation via Perturbation-Aware Contrastive Learning
Bingqian Lin, Yanxin Long, Yi Zhu, Fengda Zhu, Xiaodan Liang, Qixiang Ye, Liang Lin
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 |
|---|---|---|---|
| train with PROPER | EnvDropout · R2R Val Unseen | SR (%): 52 -> 55 | Improved |
| train with PROPER | RelGraph · R2R Val Unseen | SR (%): 57 -> 58 | Improved |
| train with PROPER | VLN BERT · R2R Test Unseen | SR (%): 63 -> 63 | No clear effect |
| add progressive trajectory augmentation to teacher+RL (ablation) | EnvDropout · R2R Val Unseen | SR (%): 47.6 -> 50.4 | Improved |
| make perturbation augmentation progressive instead of non-progressive (ablation) | EnvDropout · R2R Val Unseen | SR (%): 49.2 -> 50.4 | Improved |
| add perturbation-aware contrastive learning (ablation) | EnvDropout+Progressive Aug · R2R Val Unseen | SR (%): 50.4 -> 51.3 | Improved |
| add perturbation-aware contrastive learning (ablation) | EnvDropout+Progressive Aug · R2R Val Seen | SR (%): 62.6 -> 63.3 | Improved |
| use Teacher2Student curriculum instead of teacher forcing only (ablation) | EnvDropout · R2R Val Unseen | SR (%): 42.9 -> 45.6 | Improved |
| train with PROPER, evaluated under perturbation | EnvDropout · PP-R2R Val Unseen | SR (%): 23.9 -> 30.3 | Improved |
| train with PROPER, evaluated under perturbation | VLN BERT · PP-R2R Val Unseen | SR (%): 38.8 -> 42.9 | Improved |
| evaluate under perturbation instead of perturbation-free | RelGraph · PP-R2R Val Unseen | SR (%): 52.5 -> 27.5 | Hurt |
| add contrastive learning to perturbation-free plus perturbation-based training (ablation) | EnvDropout · PP-R2R Val Unseen | SR (%): 28.7 -> 30.3 | Improved |
Closest papers by meaning
- Let's Reward Step-by-Step: Step-Aware Contrastive Alignment for Vision-Language Navigation in Continuous Environments (ECCV 2026 · similarity 0.89)
- Contrastive Instruction-Trajectory Learning for Vision-Language Navigation (AAAI 2022 · similarity 0.89)
- DecoVLN: Decoupling Observation, Reasoning, and Correction for Vision-and-Language Navigation (CVPR 2026 · similarity 0.88)
- AgentVLN: Towards Agentic Vision-and-Language Navigation (ECCV 2026 · similarity 0.88)
- Human-Aware Vision-and-Language Navigation: Bridging Simulation to Reality with Dynamic Human Interactions (NeurIPS 2024 · similarity 0.87)
- Test-Time Adaptation for Online Vision-Language Navigation with Feedback-based Reinforcement Learning (ICML 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.
