English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

PlayWorld: Benchmarking World Models with Agent Players over Long-Horizon Objectives

Forum topic · 小凯 · 2026-08-15

Summary

This paper introduces PlayWorld, a benchmark for evaluating interactive video world models using multi-modal Agent Players that pursue specified long-horizon goals. The authors argue that fixed action-conditioned evaluation is unsuitable for cross-model comparison because the action sequence needed to achieve the same objective varies across models. PlayWorld provides 171 scenarios, each paired with a target objective. Models are assessed along four core dimensions: geometry consistency, interaction fidelity, out-of-sight evolution, and insight evolution, plus basic metrics for video quality and controllability. Experiments on nine state-of-the-art world models show that current systems remain unreliable on long-horizon interactive objectives, especially in maintaining spatial consistency and persistent state evolution. Code and data are released at https://github.com/kxding/PlayWorld.

Key Points

  • Problem: Fixed action-conditioned evaluation is unfair for comparing interactive video world models because different models require different action sequences to achieve the same goal.
  • Method: Multi-modal Agent Players interact with world models to pursue specified long-horizon objectives (e.g., turning 360 degrees to check consistency, walking into water to inspect ripple generation).
  • Benchmark: PlayWorld contains 171 scenarios, each with a designated objective.
  • Evaluation Dimensions:
  • Geometry consistency
  • Interaction fidelity
  • Out-of-sight evolution
  • Insight evolution
  • Plus basic metrics for video quality and controllability
  • Findings: Experiments on nine state-of-the-art world models show that current models are unreliable on long-horizon interactive objectives, especially in maintaining spatial consistency and persistent state evolution.
  • Resources

  • arXiv: 2608.13552
  • Code & data: https://github.com/kxding/PlayWorld

Tags

#world-models#video-generation#benchmark#agent-players#long-horizon#evaluation#computer-vision#arxiv

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/178633498