UCLA's SidewalkBench: A Sim Benchmark for Sidewalk Robot Navigation, CoRL Spotlight
zhoubolei · x · 2026-11-07
Bolei Zhou's UCLA lab and Coco Robotics present SidewalkBench, a scalable simulation benchmark for safe urban sidewalk robot navigation, accepted as a CoRL 2026 Spotlight with open-source code.
Key points:
- Built on NVIDIA Isaac Sim with GPU-accelerated rendering for standardized visual navigation evaluation
- Two scene types: 100 procedurally generated 2km×2km environments plus 11 photorealistic 3DGS-scanned real scenes
- Two-level pedestrian simulation with an SMPL-based animation pipeline delivering 60x rendering efficiency gains over prior work
- Benchmarks 9 visual navigation models across 330 unit-test, 800 pedestrian-reactive, and 105 long-horizon scenarios
- Findings: scaling sidewalk data is critical; pedestrian interaction and long-horizon robustness remain bottlenecks; synthetic data finetuning is promising
Supported by the NVIDIA Academic Grant Program.
More from Embodied
- Claude paints a watercolor tree with a robotic arm: 303 brush strokes, 4.5 hours, mixes its own colors — Successful-Earth678 · 2026-11-07
- Figure CEO Brett Adcock: 'One day the entire world will look like this' — adcock_brett · 2026-11-07
- Researcher achieves arbitrary code execution on Meta Ray-Ban smartglasses via OTA firmware extraction — evilsocket · 2026-11-07
- MACHINA Summit co-founder on Europe's Physical AI bottlenecks and VC gap — MarwaEldiwiny · 2026-11-07
- Wayve brings Wayve Labs to CoRL, booth 106, beyond-driving roadmap — Jamie_Shotton · 2026-11-07
- Claim: GPT-6 Astra's robotics skills aren't emergent — it trained on robotics-specific data — gleech · 2026-11-07