EMBODIED WEB AGENTS Benchmark
收藏资源简介:
EMBODIED WEB AGENTS Benchmark是一个包含约1500个任务的基准数据集,旨在评估人工智能体在物理和数字领域之间无缝转换的能力。该数据集涵盖了烹饪、导航、购物、旅游和地理定位等多个领域,需要人工智能体在感知、行动和网络推理之间进行协调。数据集提供了一个统一的模拟平台,集成了真实的3D室内和室外环境以及功能性的网络接口,允许人工智能体在物理和数字空间中无缝交互。该数据集可用于评估当前人工智能系统的能力,并推动未来人工智能体在集成智能领域的发展。
The EMBODIED WEB AGENTS Benchmark is a benchmark dataset comprising approximately 1,500 tasks, which aims to evaluate the capability of AI Agents to seamlessly transition between physical and digital domains. Covering multiple domains such as cooking, navigation, shopping, tourism, and geolocation, this dataset requires AI Agents to coordinate perception, action, and web reasoning. It offers a unified simulation platform that integrates realistic 3D indoor and outdoor environments alongside functional web interfaces, enabling AI Agents to interact seamlessly across both physical and digital spaces. This benchmark can be utilized to assess the capabilities of current AI systems and foster the advancement of future AI Agents in the domain of integrated intelligence.

- 1Embodied Web Agents: Bridging Physical-Digital Realms for Integrated Agent Intelligence加利福尼亚大学洛杉矶分校 · 2025年



