SidewalkBench - Gesturing: leaderboard
Navigation policies drive a simulated delivery robot past a pedestrian facing the robot with a stop gesture, where success means stopping within 3 m, in 100 procedurally generated sidewalk scenarios; the number is the success rate in percent, mean of 5 seeded runs. It isolates one pedestrian behaviour, where the long-route headline mixes them.
Metric: Success rate on 100 stop-gesture pedestrian scenarios in procedurally generated scenes, 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 | CityWalker | 11.6 | |
| 2 | S2E | 0.2 | |
| 3 | ViNT | 0 | |
| 4 | NoMaD | 0 | |
| 5 | MBRA | 0 | |
| 6 | InternVLA-N1-DualVLN | 0 | |
| 7 | MIMIC | 0 | |
| 8 | FlowPilot | 0 |
Interactive version: theaggregate.ai/benchmark?slug=sidewalkbench-gesturing · How It Works · Data refreshed daily, snapshot 2026-11-13.