通过寄存器进行动作控制
TLDR
发现世界动作模型的空间离群点,提出RegisterWAM,用可学习寄存器和流匹配监督提升机器人操作。
评分理由
The paper offers a clear analysis of representation anomalies in world action models and a novel architectural fix with strong empirical validation across multiple benchmarks and real-world tasks. However, the abstract provides limited detail on ablations, generalization limits, or comparisons to prior methods, which weakens the depth of the contribution as presented.
Read-first 评分解释
综合优先阅读分 24.5,由主题、引用、图谱、方法、可复现性和近期性等信号加权得到。 原始总分保留为 27。
研究版图角色
前沿论文
排序敏感性
稳定性:volatile;排名波动范围:195。