发表机构
Microsoft Research; DeepMind（微软研究院; 深度思考（DeepMind））
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
该研究针对GWM与RL常被混淆的问题，提出转移复杂度剖面（TCP）指标，用于量化游戏环境的转移预测难度，呼吁其成为相关论文的标准基准元数据。
AI 中文摘要
游戏世界建模（GWM）与强化学习（RL）常被混淆，因为研究论文很少量化在声明接口（带有有限历史的像素、令牌或潜变量）处的基础转移预测问题的难度。我们提出转移复杂度剖面（TCP）：一组小型、可复现的指标，通过（i）内在单步分支、（ii）可观测时的交互诱导不确定性与对手影响、（iii）通过标准化探测曲线的时空依赖跨度，来表征环境（或游戏数据集）诱导的转移核。TCP会附带明确的参考分布、协议随机性和带版本的测量预算（采样/重采样及固定探测计算），从而在各基准间实现可比数值。我们概述常见游戏族与现代“神经游戏引擎”领域如何分布于该空间，并呼吁TCP成为GWM和RL论文的标准基准元数据及必填统计量。
英文摘要
Game world modeling (GWM) and reinforcement learning (RL) are often confounded because research papers rarely quantify how difficult the underlying transition prediction problem is at the declared interface (pixels/tokens/latents with finite history). We propose the Transition Complexity Profile (TCP): a small, reproducible set of metrics that characterizes an environment's (or gameplay dataset's) induced transition kernel by (i) intrinsic one-step branching, (ii) interaction-induced uncertainty and opponent influence when observable, and (iii) temporal/spatial dependency span via standardized probe curves. TCP is reported with an explicit reference distribution, protocol stochasticity, and a versioned measurement budget (sampling/resampling and fixed probe compute), enabling comparable numbers across benchmarks. We outline how common game families and modern "neural game engine" domains populate this landscape and call for TCP to become standard benchmark metadata and a required statistic in GWM and RL papers.
CommentsAccepted by ICML 2026 Position Paper Track. https://icml.cc/virtual/2026/poster/67074
