Advertisement

Spot-Compose A Framework for Open-Vocabulary Object Retrieval and Drawer Manipulation in Point Cloud

阅读量:

文章目录

  • 概述

  • 论文解读

    • 方法架构
    • 实验
  • 代码初读

概述

accepted:Arxiv
项目主页
文章解读参考:

论文解读

在这里插入图片描述

方法架构

a comprehensive framework for robotic interaction and manipulation
3D reconstructions from a commodity 3D scanner for open-vocabulary instance segmentation , alongside grasp pose estimation

  • 三维数据的获取:iPhone 13 Pro Max using the 3D Scanner App [39]
  • Open-vocabulary 3D Instance Segmentation开放词汇3D实例分割:OpenMask3D [11]
  • 自适应抓取:AnyGrasp system [31]
  • Adaptive navigation and joint optimization
  • Dynamic drawer detection and motion estimation:三个部分drawer and handle detection, estimation of the axis of motion, and actual grasp planning
    抽屉和门把手检测 :基于RGB图像检测,YOLOv8[41], DoorDetect dataset [42]
    运动轴估计
    Refinement and impedance-based pulling

实验

Boston Dynamics Spot [37] robot
进行(1) grasping, and (2) search两个实验
inference硬件,NVIDIA RTX 4090 GPU

代码初读

项目仓库
看起来代码完整性还比较高

全部评论 (0)

还没有任何评论哟~