通过时间平直化提升潜空间规划的表示能力，让模型更稳定地达成目标。
Temporal Straightening for Latent Planning
- 引入曲率正则化使潜空间轨迹局部直线化，优化编码器与预测器联合学习
- 在多种目标导向任务中，梯度规划成功率显著提升，稳定性增强
- 适合研究世界模型、潜空间规划与强化学习的开发者和研究人员
学习良好的表示是基于世界模型进行潜空间规划的关键。尽管预训练视觉编码器能生成强大的语义视觉特征，但这些特征未针对规划任务优化，可能包含无关甚至有害于规划的信息。受人类视觉处理中感知平直化假说启发，本文提出时间平直化方法，通过曲率正则化鼓励潜空间轨迹局部直线化，联合学习一个杰帕（JEPA）世界模型的编码器与预测器。实证表明，该方法使潜空间中的欧氏距离更贴近测地距离，改善了规划目标的条件性。在一系列目标达成任务中，时间平直化显著提升了基于梯度的规划稳定性，并大幅提高成功率。代码已公开于 https://agenticlearning.ai/temporal-straightening。
原文摘要 · Abstract (English)
Learning good representations is essential for latent planning with world models. While pretrained visual encoders produce strong semantic visual features, they are not tailored to planning and contain information irrelevant -- or even detrimental -- to planning. Inspired by the perceptual straightening hypothesis in human visual processing, we introduce temporal straightening to improve representation learning for latent planning. Using a curvature regularizer that encourages locally straightened latent trajectories, we jointly learn an encoder and a predictor of a Joint-Embedding Predictive Architecture (JEPA) world model. We show that reducing curvature this way makes the Euclidean distance in latent space a better proxy for the geodesic distance and improves the conditioning of the planning objective. We demonstrate empirically that temporal straightening makes gradient-based planning more stable and yields significantly higher success rates across a suite of goal-reaching tasks. Our code is available at https://agenticlearning.ai/temporal-straightening.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。