Reproduce “SNAP: Semantic Navigation Among Pedestrians” 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
Jakob Wolfram, Pascal Roth, Fan Yang, Cesar Cadena, Marco Hutter
SNAP combines terrain categories and nearby pedestrians in a local bird’s-eye-view map, then learns a navigation policy from it. The deployed actor also uses robot motion state and goal information, but not separate pedestrian-state inputs. In a 16-pedestrian semantic-terrain simulation, 90°-view Ours* achieves 76.2% success versus ViPlanner’s 58.0%; faster pedestrians expose the 5 × 5 m map’s limited view.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.