One Scan is Enough: Demonstration-Free Adaptation for Language-Guided Navigation
One Scan is Enough: Demonstration-Free Adaptation for Language-Guided Navigation
This research proposes a method to adapt language-guided navigation policies without human demonstrations, using a single iPhone scan to create a training environment. The approach, called Scan2Instruction, synthesizes navigation episodes with grounded instructions and adapts policies through a closed-form Pareto utility. It achieves an 85% real-robot success rate across five real environments, outperforming baselines trained on human demonstrations.
This paper introduces a method for demonstration-free adaptation in language-guided navigation, using a single iPhone scan to create a training environment and adapt policies without human labels or learned critics. The approach achieves a 85% success rate in real environments, outperforming baselines trained on human demonstrations.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
complete training environment, automatically synthesizing executable navigation episodes with grounded instructions. PRPT-KS then adapts the policy through a closed-form Pareto utility read directly off the reconstructed scene, with no learned critic and no human labels. Across five real environments, the two designs together drive our RealNav to an 85% real-robot success rate, compared with 49% for a RealNav baselin
Open source S3
e post-trained on human demonstrations, and still achieve 82% in unscanned regions of the same buildings, compared with 41% for the same baseline. Our work turns demonstration-free scene adaptation for language-guided navigation from an aspiration into a working reality.
Open source S4
Source：arXiv Robotics — research abstracts · arxiv.org