SidewalkBench - Crosswalk (Real-World Scanned): leaderboard
Navigation policies drive a simulated delivery robot 10-20 m across obstacle-free crosswalks to the kerb ramp in 10 scenarios on streets scanned from the real world, without pedestrians; the number is route completion in percent, mean of 5 seeded runs. It isolates lane-following and obstacle avoidance; the pedestrian and long-route boards add people and distance.
Metric: Route completion on crosswalk blocks in real-world scanned scenes, 10 scenarios, mean of 5 seeded runs (%). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | MIMIC | 99.2 | |
| 2 | FlowPilot | 94.9 | |
| 3 | S2E | 79.3 | |
| 4 | MBRA | 65.1 | |
| 5 | NoMaD | 59.9 | |
| 6 | CityWalker | 53.7 | |
| 7 | InternVLA-N1-DualVLN | 53.1 | |
| 8 | ViNT | 37.4 |
Interactive version: theaggregate.ai/benchmark?slug=sidewalkbench-crosswalk-real-world-scanned · How It Works · Data refreshed daily, snapshot 2026-11-13.