← 研踪 PaperTrailarXiv:2606.10517 · v2
LAFP: Preserving Latent Action Structure in Latent Policy Learning via Flow Matching
首次提交 2026-06-09 15:49
版本更新 2026-11-07 21:06
雷达观测 2026-11-10 11:14
摘要
Learning high quality latent actions from large-scale unlabeled videos, coupled with limited real-world interaction data for training an action decoder, has emerged as a promising paradigm for scalable latent policy learning. However, existing approaches typically rely on behavior cloning, which tends to collapse inherently multimodal action distributions into unimodal ones, thereby degrading the pretrained latent action structure. While flow matching provides a potential alternative, directly applying it leads to a misalignment between latent actions and physical actions during action decoder training, due to the stochastic nature of the learned policy. To address these, we propose Latent Action Flow Policy (LAFP), which leverages flow matching for latent policy learning and introduces an inference-time interpolation mechanism to mitigate stochasticity-induced misalignment. Experimental results demonstrate that LAFP consistently outperforms prior methods on most downstream imitation learning tasks, achieving an 8.2\% improvement in average success rate while incurring less than 1x additional inference overhead.
作者备注：NeurIPS 2026
观测到的版本
相邻记录可能跳过中间版本；首次发现 v2 及以上也会标为版本更新。
v2 · 2026-11-07 21:06 · 最新观测版本
LAFP: Preserving Latent Action Structure in Latent Policy Learning via Flow Matching
Learning high quality latent actions from large-scale unlabeled videos, coupled with limited real-world interaction data for training an action decoder, has emerged as a promising paradigm for scalable latent policy learning. However, existing approaches typically rely on behavior cloning, which tends to collapse inherently multimodal action distributions into unimodal ones, thereby degrading the pretrained latent action structure. While flow matching provides a potential alternative, directly applying it leads to a misalignment between latent actions and physical actions during action decoder training, due to the stochastic nature of the learned policy. To address these, we propose Latent Action Flow Policy (LAFP), which leverages flow matching for latent policy learning and introduces an inference-time interpolation mechanism to mitigate stochasticity-induced misalignment. Experimental results demonstrate that LAFP consistently outperforms prior methods on most downstream imitation learning tasks, achieving an 8.2\% improvement in average success rate while incurring less than 1x additional inference overhead.
查看此版本解读 →查看此版本原文 ↗