返回全部动态
PlayWorld:用智能体玩家在长时程目标下基准测试世界模型
原标题:PlayWorld: Benchmarking World Models with Agent Players over Long-Horizon Objectives
AI 摘要
Hugging Face 每日论文介绍了 PlayWorld,这是一个用于评估交互式视频世界模型的新基准。它使用多模态智能体玩家在长时程目标下与模型交互,评估几何一致性、交互保真度和状态演化等维度。实验发现当前最先进的世界模型在长时程交互目标上仍不可靠,尤其是在空间一致性和持续状态演化方面。
以上摘要由 AI 生成,可能存在误差。事实请以原文为准。
正文节选
PlayWorld: Benchmarking World Models with Agent Players over Long-Horizon Objectives Abstract PlayWorld benchmarks interactive video world models by using multi-modal agents to pursue long-horizon objectives, evaluating geometry consistency, interaction fidelity, and state evolution. Video world models simulate future states conditioned on current observations and user actions. Recent systems have demonstrated impressive video consistency and action controllability over long sequences. However,
发布时间:—
抓取时间:2026-08-14 10:34
来源机构:Hugging Face