Reproduce “A Comprehensive Survey and Systematic Real-World Evaluation of Embodied Vision-and-Language Navigation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Liuyi Wang, Kai Sheng, Zongtao He, Jinlong Li, Yongrui Qin, Haojie Dai, Xiangyi Wang, Jingwei Yang and 3 more
A comprehensive survey and systematic real-world evaluation of embodied Vision-and-Language Navigation (VLN) systems quantify the substantial performance degradation and identify practical limitations when state-of-the-art algorithms are deployed on physical robots. The work establishes a structured taxonomy of VLN methods and pinpoints key challenges for future research.
Key insights
- There is a substantial performance gap between VLN algorithm performance in simulation and their real-world deployment on physical robots, with sim-to-real degradation being a pronounced issue.
- Hierarchical VLN frameworks, especially when coupled with panoramic sensing and SLAM, generally demonstrate better real-world transferability and safety compared to monolithic RGB-only configurations.
- Current VLN models struggle with precise semantic instruction grounding, robust collision avoidance, long-term intention tracking, and effective generalization to diverse real-world visual and physical variations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.