AUTOSIGNAL
Human-curated. Expert-annotated. Every signal traced to source.
人工精选 · 专家点评 · 每条信号可溯源
Research & IP研究与专利
Mental World Modeling: Predicting Behavior via Agents' Hidden Beliefs, Not Just Physical State心智世界建模（MWM）：让世界模型学会预测行为背后的隐藏信念
S 3.3
T1
1 sources1 个来源
R7-research
- Researchers formulate Mental World Modeling (MWM), a framework arguing that world models must track each agent's hidden mental state—beliefs, wants, intentions, feelings, and sense of social permissibility—not just the physical scene.
- The paper argues existing world models only answer a physical question (what an object is, where it is, how it will evolve), so a model that ignores what each agent knows or believes predicts the wrong action even for a correctly modeled physical scene.
- For autonomous driving, this reframing matters directly: predicting whether a pedestrian or cyclist will cross or yield depends on their belief about the vehicle's intent, not just their visible position and trajectory.
- 研究者提出心智世界模型（Mental World Modeling, MWM）框架,主张世界模型不能只建模物理场景,还必须追踪每个智能体隐藏的心智状态——信念、欲望、意图、情感以及对社会许可行为的判断。
- 论文指出,现有世界模型只回答物理层面的问题(物体是什么、在哪里、将如何演变),若不建模每个智能体知道什么、相信什么,即便物理场景预测正确,也会得出错误的行为预测。
- 对自动驾驶而言,这一转向直接相关:预测行人或骑行者是否会穿越马路或让行,取决于他们对车辆意图的信念,而不仅是其可见的位置和轨迹。