PHYRE (PHYsical REasoning)

PHYRE是物理推理的基准。它由代理必须通过执行动作来解决的物理推理任务组成。根据所需的操作类型，任务分为两层。每个层有25个任务模板，每个模板有100个任务。单击下面的任务模板以查看任务并尝试解决它们。

PHYRE is a benchmark for physical reasoning. It consists of physical reasoning tasks that agents must solve by executing actions. Tasks are divided into two tiers based on the type of required operations. Each tier contains 25 task templates, and each template has 100 tasks. Click on the task templates below to view the tasks and attempt to solve them.

搜集汇总

背景与挑战

背景概述

PHYRE是一个物理推理基准数据集，包含两层任务，每层25个任务模板，每个模板100个任务，用于评估代理的物理推理能力。该数据集由Facebook AI Research于2019年发布，适用于环境视频预测和视觉推理等研究领域。