论文揭示世界模型想象 rollout 会逐步丢失任务信号
The Posterior-Prior Epistemic Gap: When World Model Imagination Fails to Carry Task-Relevant Signals
做 model-based RL 或 active inference 规划的人该看看。它用 CARLA 里的探针实验证明想象 rollout 会悄悄丢掉障碍物信息，而且 ensemble 不确定性完全察觉不到，0.52 的 AUC 就是随机水平。这种可测量的诊断思路比单纯刷分有用。
一篇 arXiv 论文在 CARLA 驾驶模拟器中用单种子案例研究测量了 Recurrent State Space Model 世界模型的想象质量，提出"后验-先验认知鸿沟"：同一个障碍物探测探针在后验状态上 AUC 为 0.85，但应用到想象先验状态上单调衰减，15 步后降至 0.71，logit 分离度下降 57%。由于探针完全相同，衰减把信息瓶颈定位在转移模型而非表征本身。更关键的发现是规划器自身无法察觉这种失败：基于 ensemble 分歧的认知项作为障碍物分类器 AUC 仅 0.52（接近随机），属于规划时未被标记的"未知的未知"。没有特权运行时坐标时，障碍规避成功率降到 0%。
作为次要贡献，论文展示了在同一个世界模型上对 Effective Free Energy 分量做任务条件加权，可实现车道保持路线完成率 81.2 ± 1.8%，并在使用特权坐标的条件下避开全部 28 个障碍物，运行时的权重调制还支持规避后的居中回正。对做 model-based RL 或 active inference 的开发者来说，这篇论文提供了一个可复用的探针式诊断方法：不要假设想象 rollout 忠实传播任务信号，应该直接测量，并把这种衰减与规划器的认知不确定性估计区分开。
The Posterior-Prior Epistemic Gap: When World Model Imagination Fails to Carry Task-Relevant Signals
Model-based planners that roll out imagined futures assume the transition model faithfully propagates task-relevant signals---a property rarely measured directly. In a single-seed case study of a Recurrent State Space Model world model with active-inference planning in the CARLA driving simulator, we identify a \textbf{posterior-prior gap}---posterior (observation-filtered) states retain obstacle information that imagined rollout states progressively lose: a supervised obstacle probe achieves AUC 0.85 on posterior states, but the same probe applied to imagined prior states shows monotonic decay (AUC 0.85$\rightarrow$0.71 over 15 steps; logit separation drops 57\%). Because the probe is identical, the decay isolates the transition model---not the representation---as the information bottleneck. Obstacle avoidance drops to 0\% without privileged runtime coordinates, and the planner's own ensemble-based epistemic term does not flag the loss---ensemble disagreement yields AUC 0.52 as an obstacle classifier (chance level), making the failure an unflagged imagination failure---an unknown unknown at planning time. As a secondary contribution, we show that \textbf{task-conditioned weighting of Effective Free Energy components} over the same world model yields lane keeping ($81.2 \pm 1.8$\% route completion) and obstacle avoidance (28/28 avoided, \emph{using privileged coordinates}) in the CARLA simulator, with runtime weight modulation enabling post-evasion centering.
本条情报归属 AITOP「论文」核心追踪栏目。涉及核心实体与技术议题包括：CARLA、世界模型、active inference、model-based RL，读者可通过下方标签进入对应专题页查阅历史脉络与关联对比。
信源核验说明：本文原始线索采编自「arXiv cs.AI」（@Jaerock Kwon, Elahe Delavari, Donghyun Kim, Haewoon Nam），经 AITOP 多维评分管线结构化提炼。