[论文] Grasp-TAGS: Task-Aware Grasp Selection via Critic Distillation for Pre...

论文概要

研究领域: ML 作者: Xun Tu, Minghao Zou, Karthik Desingh 发布时间: 2026-11-06 arXiv: 2611.09864

中文摘要

移动操作（loco-manipulation）要求移动机械臂在长时程上协调运动和操作，是服务机器人和仓库自动化等应用中大物体重排的关键能力。尽管近期有进展，可靠执行这些任务仍然困难，因为性能依赖于复杂的机器人-物体和物体-环境交互动力学。现有方法要么通过基于模型的方法（通常需要精确动力学模型），要么通过基于学习的方法（降低了这一要求，但通常假设固定或预定义的初始抓取）。然而，不同的几何稳定抓取可能导致截然不同的下游操作性能。本文提出 Grasp-TAGS——面向目标条件大物体重排的任务感知抓取选择框架。作者在多种初始抓取配置上训练统一的 PPO 强化学习策略，并将学到的 critic 蒸馏到轻量级抓取效用网络中，该网络仅使用候选抓取配置和目标信息对候选抓取进行排序。推理时，在接触发生之前就为下游策略选择预测效用最高的抓取。实验表明，Grasp-TAGS 将重排成功率从最佳单抓取策略的 79.87% 提升到 89.84%，提高了 9.97 个百分点，并在所有评估物体上降低了最终姿态误差。

原文摘要

Loco-manipulation, which requires mobile manipulators to coordinate locomotion and manipulation over long horizons, is a key capability for large-object rearrangement in applications such as service robotics and warehouse automation. Despite recent progress, reliably executing these tasks remains challenging because performance depends on complex robot-object and object-environment interaction dynamics. Existing approaches address this challenge either through model-based methods, which often require accurate dynamics models, or through learning-based methods, which reduce this requirement but typically assume a fixed or predefined initial grasp. However, different geometrically stable grasps can lead to substantially different downstream manipulation performance. Therefore, we propose Gra...

#论文 #arXiv #ML #小凯