arXiv cs.AI· Todd Y. Zhou, Daniel Zhang·· 14 小时前AI 评分40
反事实潜在世界模型 CLWM：解决部分可观测下的反事实坍缩，提升具身推理规划成功率
Learning Counterfactual World Models for Embodied Reasoning under Partial Observability
AI 导读
研究者提出反事实潜在世界模型 CLWM，结合循环信念状态编码器、动作条件潜在动力学与对比式反事实目标，解决"反事实坍缩"——模型能预测视觉上合理的未来却无法区分行为后果不同的干预。
正文 · 原文
Abstract:World models promise a general route to embodied intelligence: learn predictive dynamics once, then reason, plan, and act with them. Increasingly, the representations beneath such models are pretrained on large-scale video, interaction, and multimodal corpora, which raises a question prediction quality alone cannot answer: when is a learned representation actually actionable? We identify a failure mode we call counterfactual collapse: a model predicts visually plausible futures while failing to distinguish interventions with different behavioral consequences. This arises whenever a representation is optimized for perceptual similarity rather than intervention structure, which is precisely the objective under which most large-scale pretrained encoders are learned. We introduce Counterfactual Latent World Models (CLWM), which combine a recurrent belief-state encoder, action-conditioned latent dynamics, and a contrastive counterfactual objective that separates futures induced by distinct interventions even when their observations look alike. Across occluded manipulation, aliased navigation, and long-horizon manipulation, CLWM improves planning success over the strongest baseline (65.1% $\to$ 74.6% on Occluded Push and 67.3% $\to$ 78.9% on Aliased Maze) and reduces exploitative planning failures (18.4% $\to$ 9.7% on Deferred Kitchen), with ablations attributing the gains to hard counterfactual negatives, especially perceptual-alias negatives. Finally, our counterfactual separability metric, which tracks planning success across the five baseline model classes ($r \ge 0.94$), is representation-agnostic: given intervention-outcome labels, it can audit any encoder, pretrained or trained from scratch, before a planner trusts it. We do not yet measure it on large-scale pretrained encoders. Here we establish the metric and its relationship to planning success for world models trained from scratch.
| Comments: | Accepted at the NeurIPS 2026 Workshop "PTA: From Pretrained Representations to Acting Agents" |
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2609.05834 [cs.AI] |
| (or arXiv:2609.05834v2 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2609.05834 arXiv-issued DOI via DataCite |
Submission history
From: Todd Zhou [view email]
[v1]
Sat, 5 Sep 2026 02:56:47 UTC (20 KB)
[v2]
Sun, 25 Oct 2026 21:12:55 UTC (20 KB)
来源：arXiv cs.AI · arxiv.org