Phy-Q
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/phy-q/benchmark
下载链接
链接失效反馈官方服务：
资源简介：
该数据集包含15个基本物理场景，旨在评估代理在物理推理智力方面的表现。该测试平台不仅旨在检验代理在局部和广泛情境下的泛化能力，还将其表现与人类玩家进行比较。评估过程中涉及多种代理，包括学习型和启发式型代理，任务是在各种场景中进行物理推理和行动选择。
This dataset comprises 15 basic physical scenarios, designed to evaluate agents' performance in physical reasoning intelligence. This testbed not only aims to examine the generalization capabilities of agents across both local and broad contexts, but also compares their performance with that of human players. A diverse set of agents, including learning-based and heuristic ones, are utilized in the evaluation pipeline, with each agent's task being to conduct physical reasoning and complete action selection across various scenarios.
