SNAP: Semantic Navigation Among Pedestrians
SNAP: Semantic Navigation Among Pedestrians
This paper introduces SNAP, a reinforcement learning navigation policy that jointly handles semantic terrain and dynamic obstacles using a single semantically annotated bird's-eye-view map. The approach avoids explicit human-state estimation and outperforms existing baselines in crowd avoidance and semantic path planning. It demonstrates zero-shot transfer to a legged robot, enabling socially compliant navigation in highly dynamic real-world environments.
The authors present a reinforcement learning policy that jointly reasons about semantic terrain and dynamic obstacles from a single annotated bird's-eye-view map, achieving zero-shot transfer to a legged robot in real-world crowded scenarios without explicit human-state estimation.
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
- code: not_reported
- weights: not_reported
- dataset: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
Source：arXiv Robotics — research abstracts · arxiv.org