arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2503.06300cs.RO

接触因子图中面向操控规划的高效梯度推理

Efficient Gradient-Based Inference for Manipulation Planning in Contact Factor Graphs

Seoul National University(首尔大学)

Jeongmin Lee, Sunkyung Park, Minji Lee, Dongjun Lee

AI总结:

本文提出接触因子图(CFG)建模接触与动力学约束,并利用可微因子和结构化条件概率开发高效推理算法,实现操控规划中的分布近似与可行解采样。

摘要:

本文提出了一个框架,旨在解决操控中出现的一系列规划问题,这些问题通常涉及与接触和动力学约束相关的复杂几何-物理推理。我们引入接触因子图(Contact Factor Graph,CFG)来对这些多样化的因子进行图形化建模,使我们能够在图上执行推理,以近似分布并采样合适的解。我们提出了一种新方法,可以将接触操控的各种现象作为可微因子纳入其中,并为CFG开发了一种高效的推理算法,该算法利用了这种可微性以及由接触的结构化特性产生的条件概率。我们的结果证明了该框架在各种操控场景中生成可行样本和近似后验分布的能力。

英文摘要:

This paper presents a framework designed to tackle a range of planning problems arise in manipulation, which typically involve complex geometric-physical reasoning related to contact and dynamic constraints. We introduce the Contact Factor Graph (CFG) to graphically model these diverse factors, enabling us to perform inference on the graphs to approximate the distribution and sample appropriate solutions. We propose a novel approach that can incorporate various phenomena of contact manipulation as differentiable factors, and develop an efficient inference algorithm for CFG that leverages this differentiability along with the conditional probabilities arising from the structured nature of contact. Our results demonstrate the capability of our framework in generating viable samples and approximating posterior distributions for various manipulation scenarios.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑