360CityArena：面向具身智能体的真实城市导航基准

AI 摘要

360CityArena 是一个用于评估具身智能体城市探索能力的基准，基于东京秋叶原地区的 602 段 360 度视频构建，覆盖 85 条街道，包含 175 个人工设计的任务，涵盖环境理解、路径推理和空间推理三类。评估显示，最强的 LMM 智能体 Gemini 2.5 Flash 准确率仅 17.1%，远低于人类的 77.3%，揭示了城市级具身导航和推理仍面临巨大挑战。该基准为真实城市区域的导航和空间推理提供了必要的测试平台。

正文节选

360CityArena: A Realistic Virtual Urban Navigation Benchmark for Embodied Agents Abstract A new photorealistic urban benchmark reveals large performance gaps for embodied agents in city-scale navigation and spatial reasoning. We present 360CityArena, a benchmark for evaluating the urban exploration capabilities of embodied agents within a photorealistic environment constructed from 360-degree videos. Existing outdoor benchmarks either lack sufficient photorealism or complexity, resulting in a co

发布时间：—

抓取时间：2026-08-12 16:22

来源机构：Hugging Face