SidewalkBench - Long-Horizon (Real-World Scanned): leaderboard
The same long-route test on 5 sidewalk routes scanned from real streets (3D Gaussian splatting): failures (collisions, lane exits and freezes) per 100 m travelled, lower is better. The procedurally generated routes are the headline board.
Metric: Failures per 100 m (collisions, lane exits and freezes) on 5 long-horizon routes in real-world scanned scenes (lower is better). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | S2E | 0.92 | |
| 2 | FlowPilot | 0.92 | |
| 3 | MIMIC | 1.66 | |
| 4 | MBRA | 5.15 | |
| 5 | NoMaD | 5.89 | |
| 6 | InternVLA-N1-DualVLN | 9.39 | |
| 7 | CityWalker | 10.86 | |
| 8 | ViNT | 24.11 |
Interactive version: theaggregate.ai/benchmark?slug=sidewalkbench-long-horizon-real-world-scanned · How It Works · Data refreshed daily, snapshot 2026-11-13.