SidewalkBench is a new benchmark for visual navigation on urban sidewalks, built on NVIDIA Isaac Sim. It provides GPU-accelerated, high-fidelity environments that mix procedurally generated scenes with real-world scanned ones, and populates them with reactive, event-based pedestrian behaviors.

The authors evaluated nine visual navigation models on 330 unit-test scenarios, 800 pedestrian-reactive scenarios, and 105 long-horizon scenarios. Their results indicate that pedestrian interaction and long-horizon robustness remain critical bottlenecks for current models.

The paper suggests that scaling up sidewalk training with synthetic data is a promising direction. Because the benchmark is built on a simulator, it enables standardized, reproducible evaluation under realistic conditions.