想像的展開是運動學的，而非動力學的：長時域世界模型失敗的診斷

摘要

長期視域下的世界模型失敗通常被歸因於誤差累積，但這種泛泛的表述並未區分是哪一類誤差在累積。我們提出一個「運動學 vs 動力學」的重構框架：世界模型傾向於以運動學而非動力學的方式進行想像。我們將其操作化為「想像運動學一致性誤差」（iKCE），這是一種每步診斷指標，用於衡量模型推演偏離封閉式運動學零假設的程度；同時搭配一個擾動實驗方案，用以檢驗當物理條件跨過相態邊界時 iKCE 是否會產生響應。我們在已發布的、於 DMC walker-walk 任務上訓練的 DreamerV3 檢查點上實例化此診斷，結果顯示模型想像出的 iKCE 約比匹配的真實物理推演高出兩個數量級。在一組跨越步態崩潰邊界的摩擦係數掃描中，即便訓練策略的回報在同一範圍內急遽下降，模型的 iKCE 在統計上仍保持平穩，這正是「運動學而非動力學」的特徵。該診斷能在時間跨度超過軀體步態週期時，區分運動學想像與動力學想像。

English

Long-horizon failure in world models is conventionally attributed to compounding error, a generic framing that does not distinguish what kind of error compounds. We propose a kinematic-vs-dynamic reframing: world models tend to imagine kinematically rather than dynamically. We operationalize this as the imagined Kinematic-Consistency Error, a per-step diagnostic that measures how far a rollout departs from a closed-form kinematic null, paired with a perturbation protocol that tests whether iKCE responds when physical conditions cross a regime boundary. We instantiate the diagnostic on a released DreamerV3 checkpoint trained on DMC walker-walk, where imagined iKCE runs roughly two orders of magnitude above that of matched real-physics rollouts. Across a friction sweep that crosses the gait-collapse boundary, the model's iKCE stays statistically flat even as the trained policy's reward collapses through the same range, providing the kinematic-not-dynamic signature. The diagnostic distinguishes kinematic from dynamic imagination at horizons longer than the embodiment's gait period.