SG-WAM：几何感知策略空间中的自引导世界建模
原标题：SG-WAM: Self-Guided World Modeling in Geometry-Aware Policy Space
AI 摘要
SG-WAM 是一种自引导世界建模框架，在几何感知的策略空间中学习动作条件动力学。该框架通过可学习的动力学令牌和自引导世界预测器，在策略表示空间中预测未来状态，并结合几何监督和流匹配动作生成进行端到端优化。基于 0.9B 参数的模型，SG-WAM 在 LIBERO 基准上达到 98.5% 的平均成功率，在 LIBERO-Plus 上达到 73%，无需大规模具身预训练，且在分布内和分布外真实世界评估中均优于强基线。
正文节选
SG-WAM: Self-Guided World Modeling in Geometry-Aware Policy Space Abstract World Action Models (WAMs) couple action generation with prediction of future states. Their effectiveness depends on whether future dynamics are modeled in a space that is both aligned with action generation and sufficiently geometry-aware to capture where and how actions change the scene. Existing WAMs typically satisfy only part of this requirement, relying on either perceptually heavy observation-space targets or auxil