通用抓取位姿
2024-08-09 11:25:08 0 举报AI智能生成
通用物体抓取思路拆解
作者其他创作
大纲/内容
graspness measure
point-wise graspness scores & view-wise graspness scores
point-wise graspness scores & view-wise graspness scores
Single Object Graspness
采样点选取
夹爪深度
夹爪旋转
force analytic mode
公式
c有效抓取位姿阈值（手动设定）
Scene-Level Graspness
问题1：抓取位姿与背景物或者临近物体干涉
GraspNet-1Billion: A Large-Scale Benchmark for General Object Grasping
reconstruct the scene using object 3D models and correspond-
ing 6D poses
ing 6D poses
碰撞检测
公式
Ckij-collision label
问题2：RGBD观测到的是单视野局部点云
associate the scene point cloud with the projected object point
通过对象的6D Pose将对象点投影到场景中
公式
归一化表示
cascaded graspness model
作用：提取每个点的局部特征向量
局部特征向量：
作用：提取每个点的局部特征向量
局部特征向量：
Two sub-function
F - 高维特征
point level过滤掉大部分不可抓取点；View-wise只需要计算剩下的点
backbone network
ResUNet14
extraction of both global and
local point features
extraction of both global and
local point features
功能
Graspable Farthest Point Sampling
功能
使用MLP建模
to generate point-wise graspable landscape.
to generate point-wise graspable landscape.
select points with graspness score larger than δp
adopt farthest point sampling (FPS) to maximize distances among
sampled points
sampled points
Graspable Probabilistic View Selection
功能
使用MLP建模
to the sampled seed points and output M × V vectors for view-wise graspable landscapes
and M × C residual features for grasp generation.
to the sampled seed points and output M × V vectors for view-wise graspable landscapes
and M × C residual features for grasp generation.
Fibonacci lattices生成V个观测视野
grasp operation model
Crop-and-refine
从点云中裁剪出潜在的抓取区域，然后对这些区域进行细化处理，以精确估计抓取位姿和提高抓取检测的准确性。通过这种方式，可以从复杂的场景中有效地识别和定位可抓取的物体。
从点云中裁剪出潜在的抓取区域，然后对这些区域进行细化处理，以精确估计抓取位姿和提高抓取检测的准确性。通过这种方式，可以从复杂的场景中有效地识别和定位可抓取的物体。
Cylinder-Grouping from Seed Points
Grasp Generation from Candidates
PointNet
for grasp generation
for grasp generation
Grasp Score Representation
minimum friction coefficient
Loss Function
Loss Function说明
0 条评论
下一页
