Reproduce “SidewalkBench: Benchmarking Visual Navigation on Urban Sidewalks” 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
Zhizheng Liu, Honglin He, Vivek Alumootil, Akshat Pandya, Brad Squicciarini, Wayne Wu, Bolei Zhou
SidewalkBench is a benchmark for evaluating sidewalk navigation across procedural and scanned scenes. It tests nine visual-navigation models on basic layouts, eight pedestrian behaviors, and routes longer than 100 m. Pedestrian-crossing success averages 3.8% across the models. A small real-world test finds FlowPilot crossing success rises from 0% to 40%, but does not establish broad sim-to-real transfer.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.