預測動力學能否存在於物理世界中？

摘要

預測性物理AI系統會輸出狀態展開、動作塊與潛在計劃，然而低均方根誤差（RMSE）並不代表某一特定提案在物理上可執行。我們將物理可接受性定義為預測-控制介面：在執行前，經解碼的提案被視為候選動力學模型，並透過運動學、動力學及直接至複合時域條件進行評估。通過並不代表任務成功；被拒絕則表示違反了指定的物理範圍，並會提供組成層級的理由。在Hugging Face LeRobot PushT上進行的控制性偽造測試顯示：單步預測RMSE與標準化動力學殘差的ROC曲線下面積（AUC）分別達到0.982與0.972，僅使用運動學條件則AUC為0.592，而完整閘控系統在具備條件層級歸因下AUC達到0.957。在基於重播的干預實驗中，殘差型濾波器與完整物理可接受性閘控能阻止87%至89%的無效提案，同時維持平均進度接近0.998。

English

Predictive Physical AI systems output state rollouts, action chunks, and latent plans, yet a low root-mean-square error (RMSE) does not imply that a particular proposal is physically executable. We formulate physical admissibility as a prediction-control interface: before execution, a decoded proposal is treated as candidate dynamics and evaluated using kinematic, dynamic, and direct-to-composed horizon conditions. Passing is not a certificate of task success; rejection identifies violation of the specified physical envelope and gives a component-level reason. On Hugging Face LeRobot PushT, controlled falsification shows that one-step prediction-RMSE and standardized dynamics residuals reach area under the receiver operating characteristic curve (AUC) 0.982 and 0.972, kinematic-only conditions reach AUC 0.592, and the full gate reaches AUC 0.957 with condition-level attribution. In replay-based intervention experiments, residual-based filters and the full physical-admissibility gate prevent 87-$89% of invalid proposals while preserving mean progress near 0.998.